diff --git a/changelog.d/fixes/vertex-express-xai-models.md b/changelog.d/fixes/vertex-express-xai-models.md
new file mode 100644
index 00000000000..5752468279a
--- /dev/null
+++ b/changelog.d/fixes/vertex-express-xai-models.md
@@ -0,0 +1 @@
+- **fix(vertex):** prefer live model discovery for OAuth, Service Account, and service-account-bound authorization-key credentials while falling back cleanly to the Gemini-only Express catalog for standard API keys. Model Garden resources now route by publisher protocol: Claude and Mistral use their native `rawPredict` APIs, and Grok plus current or future open MaaS publishers use Vertex's OpenAI-compatible endpoint with normalized request IDs such as `xai/grok-4.6`. Distinguish intentional API-key catalog rejection from transient HTTP or network failures. ([#12471](https://github.com/diegosouzapw/OmniRoute/pull/12471)) — thanks @JxnLexn
diff --git a/next.config.mjs b/next.config.mjs
index f4d4622c90b..15085232807 100644
--- a/next.config.mjs
+++ b/next.config.mjs
@@ -349,6 +349,10 @@ const nextConfig = {
     "keytar",
     "wreq-js",
     "zod",
+    // jsdom relies on Node class relationships that Turbopack's server-chunk transform can break
+    // (observed as "Class extends value undefined" during Vertex metadata sync). Keep the native
+    // package boundary; standalone file tracing still copies the runtime dependency.
+    "jsdom",
     "@ngrok/ngrok",
     "@huggingface/transformers",
     // The ESM entry imports tiktoken_bg.wasm as a module. Turbopack can compile
diff --git a/open-sse/config/freeModelCatalog.data.ts b/open-sse/config/freeModelCatalog.data.ts
index 3d7ebed5950..cc934744c32 100644
--- a/open-sse/config/freeModelCatalog.data.ts
+++ b/open-sse/config/freeModelCatalog.data.ts
@@ -442,11 +442,17 @@ export const FREE_MODEL_BUDGETS: FreeModelBudget[] = [
   { provider: "vertex", modelId: "gemini-3.1-pro-preview", displayName: "Gemini 3.1 Pro Preview (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
   { provider: "vertex", modelId: "gemini-3.1-flash-lite", displayName: "Gemini 3.1 Flash Lite (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
   { provider: "vertex", modelId: "gemini-3-flash-preview", displayName: "Gemini 3 Flash Preview (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
-  { provider: "vertex", modelId: "gemma-4-31b-it", displayName: "Gemma 4 31B (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
-  { provider: "vertex", modelId: "DeepSeek-V4-Flash", displayName: "DeepSeek V4 Flash (Vertex Partner)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
-  { provider: "vertex", modelId: "DeepSeek-V4-Pro", displayName: "DeepSeek V4 Pro (Vertex Partner)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
-  { provider: "vertex", modelId: "Qwen3.6-35B-A3B", displayName: "Qwen3.6 35B A3B (Vertex Partner)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
-  { provider: "vertex", modelId: "GLM-5.1-FP8", displayName: "GLM-5.1 (Vertex Partner)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
+  { provider: "vertex", modelId: "gemini-3.7-flash", displayName: "Gemini 3.7 Flash (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
+  { provider: "vertex", modelId: "gemini-3.6-flash", displayName: "Gemini 3.6 Flash (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
+  { provider: "vertex", modelId: "gemini-3.5-flash", displayName: "Gemini 3.5 Flash (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
+  { provider: "vertex", modelId: "gemini-3.5-flash-lite", displayName: "Gemini 3.5 Flash-Lite (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
+  { provider: "vertex", modelId: "gemini-2.5-pro", displayName: "Gemini 2.5 Pro (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
+  { provider: "vertex", modelId: "gemini-2.5-flash", displayName: "Gemini 2.5 Flash (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
+  { provider: "vertex", modelId: "gemini-2.5-flash-lite", displayName: "Gemini 2.5 Flash-Lite (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
+  { provider: "vertex", modelId: "deepseek-ai/deepseek-v3.2-maas", displayName: "DeepSeek V3.2 (Vertex MaaS)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
+  { provider: "vertex", modelId: "deepseek-ai/deepseek-v3.1-maas", displayName: "DeepSeek V3.1 (Vertex MaaS)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
+  { provider: "vertex", modelId: "qwen/qwen3-next-80b-a3b-instruct-maas", displayName: "Qwen3 Next 80B Instruct (Vertex MaaS)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
+  { provider: "vertex", modelId: "zai-org/glm-5-maas", displayName: "GLM 5 (Vertex MaaS)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
   { provider: "vertex", modelId: "claude-opus-4-7", displayName: "Claude Opus 4.7 (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
   { provider: "vertex", modelId: "claude-sonnet-4-6", displayName: "Claude Sonnet 4.6 (Vertex)", monthlyTokens: 0, creditTokens: 300000000, freeType: "one-time-initial", poolKey: "vertex", tos: "caution" },
   { provider: "requesty", modelId: "openai/gpt-oss-120b", displayName: "GPT-OSS 120B (Requesty free)", monthlyTokens: 0, creditTokens: 0, freeType: "recurring-uncapped", poolKey: "requesty-free", tos: "ok" },
diff --git a/open-sse/config/providerModels.ts b/open-sse/config/providerModels.ts
index 3e6dffbf99e..5f6eb189d98 100644
--- a/open-sse/config/providerModels.ts
+++ b/open-sse/config/providerModels.ts
@@ -1,4 +1,5 @@
 import { generateModels, generateAliasMap, type RegistryModel } from "./providerRegistry.ts";
+import { getVertexModelTargetFormat } from "./vertexModels.ts";
 
 // Lazy PROVIDER_MODELS: deferred until first property access to speed up startup.
 // The Proxy defers `generateModels()` from module-evaluation time to the first read.
@@ -228,9 +229,10 @@ export function getModelTargetFormat(aliasOrId: string, modelId: string): string
   // executor's /codex/i routing, 9router#102). Scoped to the openai alias so other
   // providers shipping *-pro ids keep their own endpoint semantics.
   if (alias === "openai" && /-pro$/i.test(bareModelId)) return "openai-responses";
-  // ponytail: Claude models on Vertex use rawPredict with Anthropic Messages format,
-  // not the Gemini generateContent format. Mirrors executor isClaudeModel() check.
-  if ((alias === "vertex" || alias === "vp") && /^claude-/i.test(bareModelId)) return "claude";
+  // Vertex uses three protocol families: Gemini generateContent, Anthropic Messages rawPredict,
+  // and OpenAI-shaped Mistral/Open-MaaS requests. Resource names retain enough publisher data to
+  // route future dynamically-synced models without adding another pinned prefix here.
+  if (alias === "vertex" || alias === "vp") return getVertexModelTargetFormat(bareModelId);
   // Model-level targetFormat is provider-scoped: a catalog entry declares how THIS
   // provider's endpoint serves the model — do NOT import another provider's tag.
   // #9994 scoped this for providers WITH a catalog; #10072 extends it to catalogless
diff --git a/open-sse/config/providers/registry/vertex/index.ts b/open-sse/config/providers/registry/vertex/index.ts
index 4b72b9f7c45..1a8975e2e30 100644
--- a/open-sse/config/providers/registry/vertex/index.ts
+++ b/open-sse/config/providers/registry/vertex/index.ts
@@ -1,4 +1,5 @@
 import type { RegistryEntry } from "../../shared.ts";
+import { VERTEX_XAI_MODELS } from "../../../vertexModels.ts";
 
 export const vertexProvider: RegistryEntry = {
   id: "vertex",
@@ -9,7 +10,7 @@ export const vertexProvider: RegistryEntry = {
   // URL uses {project_id} and {region} from providerSpecificData — handled by custom executor or fallback
   // Default to us-central1 / generic endpoint; users configure project via providerSpecificData
   baseUrl: "https://us-central1-aiplatform.googleapis.com/v1/projects",
-  urlBuilder: (base, model, stream) => {
+  urlBuilder: (_base, model, stream) => {
     // Full URL: {base}/{project}/locations/{region}/publishers/google/models/{model}:{action}
     // For a generic fallback, we build a Gemini-compatible URL
     // The actual project/region are configured via providerSpecificData in the DB connection
@@ -19,14 +20,33 @@ export const vertexProvider: RegistryEntry = {
   authType: "apikey",
   authHeader: "bearer",
   models: [
+    { id: "gemini-3.7-flash", name: "Gemini 3.7 Flash (Vertex)" },
+    { id: "gemini-3.6-flash", name: "Gemini 3.6 Flash (Vertex)" },
+    { id: "gemini-3.5-flash", name: "Gemini 3.5 Flash (Vertex)" },
+    { id: "gemini-3.5-flash-lite", name: "Gemini 3.5 Flash-Lite (Vertex)" },
     { id: "gemini-3.1-pro-preview", name: "Gemini 3.1 Pro Preview (Vertex)" },
     { id: "gemini-3.1-flash-lite", name: "Gemini 3.1 Flash Lite (Vertex)" },
     { id: "gemini-3-flash-preview", name: "Gemini 3 Flash Preview (Vertex)" },
-    { id: "gemma-4-31b-it", name: "Gemma 4 31B (Vertex)" },
-    { id: "DeepSeek-V4-Flash", name: "DeepSeek V4 Flash (Vertex Partner)" },
-    { id: "DeepSeek-V4-Pro", name: "DeepSeek V4 Pro (Vertex Partner)" },
-    { id: "Qwen3.6-35B-A3B", name: "Qwen3.6 35B A3B (Vertex Partner)" },
-    { id: "GLM-5.1-FP8", name: "GLM-5.1 (Vertex Partner)" },
+    { id: "gemini-2.5-pro", name: "Gemini 2.5 Pro (Vertex)" },
+    { id: "gemini-2.5-flash", name: "Gemini 2.5 Flash (Vertex)" },
+    { id: "gemini-2.5-flash-lite", name: "Gemini 2.5 Flash-Lite (Vertex)" },
+    {
+      id: "deepseek-ai/deepseek-v3.2-maas",
+      name: "DeepSeek V3.2 (Vertex MaaS)",
+      targetFormat: "openai",
+    },
+    {
+      id: "deepseek-ai/deepseek-v3.1-maas",
+      name: "DeepSeek V3.1 (Vertex MaaS)",
+      targetFormat: "openai",
+    },
+    {
+      id: "qwen/qwen3-next-80b-a3b-instruct-maas",
+      name: "Qwen3 Next 80B Instruct (Vertex MaaS)",
+      targetFormat: "openai",
+    },
+    { id: "zai-org/glm-5-maas", name: "GLM 5 (Vertex MaaS)", targetFormat: "openai" },
+    ...VERTEX_XAI_MODELS,
     { id: "claude-fable-5-1", name: "Claude Fable 5.1 (Vertex)", targetFormat: "claude" },
     { id: "claude-fable-5", name: "Claude Fable 5 (Vertex)", targetFormat: "claude" },
     { id: "claude-opus-5", name: "Claude Opus 5 (Vertex)", targetFormat: "claude" },
@@ -41,4 +61,7 @@ export const vertexProvider: RegistryEntry = {
     { id: "claude-haiku-4-5", name: "Claude Haiku 4.5 (Vertex)", targetFormat: "claude" },
   ],
   passthroughModels: true,
+  // Gemini + publisher discovery are independent APIs. A partial successful response must not
+  // hide curated partner models omitted by another publisher catalog.
+  liveCatalogAuthoritative: false,
 };
diff --git a/open-sse/config/providers/registry/vertex/partner/index.ts b/open-sse/config/providers/registry/vertex/partner/index.ts
index 6524cd6a2b2..0ce2be94ac6 100644
--- a/open-sse/config/providers/registry/vertex/partner/index.ts
+++ b/open-sse/config/providers/registry/vertex/partner/index.ts
@@ -1,4 +1,5 @@
 import type { RegistryEntry } from "../../../shared.ts";
+import { VERTEX_XAI_MODELS } from "../../../../vertexModels.ts";
 
 export const vertex_partnerProvider: RegistryEntry = {
   id: "vertex-partner",
@@ -9,10 +10,15 @@ export const vertex_partnerProvider: RegistryEntry = {
   authType: "apikey",
   authHeader: "bearer",
   models: [
-    { id: "DeepSeek-V4-Flash", name: "DeepSeek V4 Flash" },
-    { id: "DeepSeek-V4-Pro", name: "DeepSeek V4 Pro" },
-    { id: "Qwen3.6-35B-A3B", name: "Qwen 3.6 35B A3B" },
-    { id: "GLM-5.1-FP8", name: "GLM 5.1" },
+    { id: "deepseek-ai/deepseek-v3.2-maas", name: "DeepSeek V3.2", targetFormat: "openai" },
+    { id: "deepseek-ai/deepseek-v3.1-maas", name: "DeepSeek V3.1", targetFormat: "openai" },
+    {
+      id: "qwen/qwen3-next-80b-a3b-instruct-maas",
+      name: "Qwen3 Next 80B Instruct",
+      targetFormat: "openai",
+    },
+    { id: "zai-org/glm-5-maas", name: "GLM 5", targetFormat: "openai" },
+    ...VERTEX_XAI_MODELS,
     { id: "claude-fable-5-1", name: "Claude Fable 5.1", targetFormat: "claude" },
     { id: "claude-fable-5", name: "Claude Fable 5", targetFormat: "claude" },
     { id: "claude-opus-5", name: "Claude Opus 5", targetFormat: "claude" },
@@ -27,4 +33,5 @@ export const vertex_partnerProvider: RegistryEntry = {
     { id: "claude-opus-4-5", name: "Claude Opus 4.5", targetFormat: "claude" },
     { id: "claude-haiku-4-5", name: "Claude Haiku 4.5", targetFormat: "claude" },
   ],
+  liveCatalogAuthoritative: false,
 };
diff --git a/open-sse/config/vertexModels.ts b/open-sse/config/vertexModels.ts
new file mode 100644
index 00000000000..448c23eb525
--- /dev/null
+++ b/open-sse/config/vertexModels.ts
@@ -0,0 +1,106 @@
+import type { RegistryModel } from "./providers/shared.ts";
+
+const VERTEX_PUBLISHER_RESOURCE_PATTERN =
+  /^(?:(?:projects\/[^/]+\/locations\/[^/]+\/)?publishers\/([^/]+)\/models\/|([^/]+)\/models\/)(.+)$/i;
+const NATIVE_PUBLISHERS = new Set(["google", "anthropic", "mistralai"]);
+
+const LEGACY_OPENAI_MAAS_PUBLISHERS = [
+  { prefix: "deepseek-", publisher: "deepseek-ai" },
+  { prefix: "qwen", publisher: "qwen" },
+  { prefix: "llama-", publisher: "meta" },
+  { prefix: "glm-", publisher: "zai-org" },
+  { prefix: "gpt-oss-", publisher: "openai" },
+  { prefix: "kimi-", publisher: "moonshotai" },
+  { prefix: "minimax-", publisher: "minimaxai" },
+] as const;
+
+export type VertexModelTransport = "gemini" | "anthropic" | "mistral" | "openai";
+
+function normalizePublisherModel(publisher: string, model: string): string {
+  const normalizedPublisher = publisher.toLowerCase();
+  if (NATIVE_PUBLISHERS.has(normalizedPublisher)) return model;
+  return `${normalizedPublisher}/${model}`;
+}
+
+/**
+ * Convert Model Garden resource/version names to the model value accepted by its inference API.
+ * Native Google, Anthropic, and Mistral endpoints encode the publisher in the URL and take a bare
+ * model id. OpenAI-compatible MaaS requests keep the publisher namespace in the request body.
+ */
+export function normalizeVertexModelId(model: string): string {
+  const trimmed = model.trim().replace(/^(?:vertex|vp)\//i, "");
+  const resourceMatch = trimmed.match(VERTEX_PUBLISHER_RESOURCE_PATTERN);
+  if (resourceMatch) {
+    const publisher = resourceMatch[1] || resourceMatch[2];
+    const modelId = resourceMatch[3];
+    if (publisher && modelId) return normalizePublisherModel(publisher, modelId);
+  }
+
+  const namespacedMatch = trimmed.match(/^([^/]+)\/(.+)$/);
+  if (namespacedMatch && NATIVE_PUBLISHERS.has(namespacedMatch[1].toLowerCase())) {
+    return namespacedMatch[2];
+  }
+
+  // Keep manually-added historical Grok ids working while emitting Google's documented MaaS
+  // request form (`xai/grok-*`) for the OpenAI-compatible endpoint.
+  if (/^grok-/i.test(trimmed)) return `xai/${trimmed}`;
+  const legacyMaaS = LEGACY_OPENAI_MAAS_PUBLISHERS.find(({ prefix }) =>
+    trimmed.toLowerCase().startsWith(prefix)
+  );
+  if (legacyMaaS) return `${legacyMaaS.publisher}/${trimmed}`;
+  return trimmed;
+}
+
+/** Resolve the Vertex transport from either a request id or a Model Garden resource name. */
+export function getVertexModelTransport(model: string): VertexModelTransport {
+  const normalized = normalizeVertexModelId(model).toLowerCase();
+  if (normalized.startsWith("claude-")) return "anthropic";
+  if (normalized.startsWith("mistral-")) return "mistral";
+  if (normalized.includes("/")) return "openai";
+  return "gemini";
+}
+
+/** Wire format used before the request reaches the provider-specific Vertex executor. */
+export function getVertexModelTargetFormat(model: string): "claude" | "openai" | null {
+  const transport = getVertexModelTransport(model);
+  if (transport === "anthropic") return "claude";
+  if (transport === "mistral" || transport === "openai") return "openai";
+  return null;
+}
+
+/** True for xAI models served through Vertex's OpenAI-compatible MaaS endpoint. */
+export function isVertexXaiModel(model: string): boolean {
+  return /^xai\/grok-/i.test(normalizeVertexModelId(model));
+}
+
+/** True only for models exposed by the project-less Vertex Express API. */
+export function isVertexExpressModel(model: string): boolean {
+  return (
+    getVertexModelTransport(model) === "gemini" && /^gemini-/i.test(normalizeVertexModelId(model))
+  );
+}
+
+/** Curated fallback metadata for Vertex xAI MaaS models. */
+export const VERTEX_XAI_MODELS = [
+  {
+    id: "xai/grok-4.6",
+    name: "Grok 4.6",
+    contextLength: 524288,
+    supportsReasoning: true,
+    supportsVision: true,
+    toolCalling: true,
+    targetFormat: "openai",
+  },
+  { id: "xai/grok-4.3", name: "Grok 4.3", targetFormat: "openai" },
+  {
+    id: "xai/grok-4.20-reasoning",
+    name: "Grok 4.20 Reasoning",
+    supportsReasoning: true,
+    targetFormat: "openai",
+  },
+  {
+    id: "xai/grok-4.20-non-reasoning",
+    name: "Grok 4.20 Non-Reasoning",
+    targetFormat: "openai",
+  },
+] as const satisfies readonly RegistryModel[];
diff --git a/open-sse/executors/vertex.ts b/open-sse/executors/vertex.ts
index 9d9ed6502f2..80fc0c78791 100644
--- a/open-sse/executors/vertex.ts
+++ b/open-sse/executors/vertex.ts
@@ -1,6 +1,7 @@
 import { SignJWT, importPKCS8 } from "jose";
 import { BaseExecutor, ExecuteInput } from "./base.ts";
 import { PROVIDERS } from "../config/constants.ts";
+import { getVertexModelTransport, normalizeVertexModelId } from "../config/vertexModels.ts";
 
 interface ServiceAccount {
   type: string;
@@ -26,18 +27,25 @@ export const VERTEX_OAUTH_SCOPES = [
 ] as const;
 
 export function parseSAFromApiKey(apiKey: string): ServiceAccount {
+  let parsed: ServiceAccount & { web?: unknown; installed?: unknown };
   try {
-    return JSON.parse(apiKey);
+    parsed = JSON.parse(apiKey);
   } catch {
     throw new Error("Vertex AI requires a valid Service Account JSON as the API key");
   }
+  if (parsed && typeof parsed === "object" && ("web" in parsed || "installed" in parsed)) {
+    throw new Error(
+      "OAuth client JSON is not a Service Account key; create a Service Account JSON key or provide an OAuth access token"
+    );
+  }
+  return parsed;
 }
 
 /**
  * A Service Account credential is a JSON object (type/client_email/private_key). A Vertex AI
  * Express-mode API key is an opaque non-JSON string. Distinguishing them lets the executor
  * support BOTH: Service Account JSON (JWT → OAuth → project-scoped endpoint + Bearer auth) and
- * Express keys (project-less publisher endpoint + x-goog-api-key auth), instead of failing every
+ * Express keys (project-less publisher endpoint + query-key auth), instead of failing every
  * Express key with "requires a valid Service Account JSON".
  */
 export function looksLikeServiceAccountJson(apiKey: string): boolean {
@@ -52,7 +60,9 @@ export function looksLikeServiceAccountJson(apiKey: string): boolean {
 
 /** True for a Vertex AI Express-mode API key (a non-empty, non-JSON, non-OAuth credential). */
 export function isExpressApiKey(apiKey?: string | null): boolean {
-  return typeof apiKey === "string" && apiKey.trim().length > 0 && !looksLikeServiceAccountJson(apiKey);
+  return (
+    typeof apiKey === "string" && apiKey.trim().length > 0 && !looksLikeServiceAccountJson(apiKey)
+  );
 }
 
 export async function getAccessToken(sa: ServiceAccount): Promise<string> {
@@ -115,35 +125,97 @@ export async function getAccessToken(sa: ServiceAccount): Promise<string> {
   return accessToken;
 }
 
-const PARTNER_MODELS = new Set([
-  // Generic prefix, not pinned to a version: every Claude model on Vertex is an Anthropic
-  // partner model, never a Google-publisher one. Pinned prefixes (e.g. "claude-3-5-sonnet")
-  // silently break every time Anthropic ships a new generation — see issue #1985.
-  "claude-",
-  "deepseek-v3",
-  "deepseek-v3.2",
-  "deepseek-v4",
-  "deepseek-deepseek-r1",
-  "qwen3-next-80b",
-  "qwen3.6-",
-  "llama-3.1",
-  "mistral-",
-  "glm-5",
-  "glm-5.1",
-  "meta/llama",
-]);
-
 function isPartnerModel(model: string) {
-  const normalizedModel = model.toLowerCase();
-  return [...PARTNER_MODELS].some((prefix) => normalizedModel.startsWith(prefix));
+  return getVertexModelTransport(model) === "openai";
 }
 
 // Anthropic models need their own branch: they use Vertex's native Anthropic Messages API
-// (publishers/anthropic/.../rawPredict), not the generic OpenAI-compatible partner endpoint the
-// other PARTNER_MODELS entries (DeepSeek, Qwen, Llama, Mistral, GLM) go through — the OpenAI-shaped
-// endpoint 404s/"malformed argument"s for Claude models on at least some projects.
+// (publishers/anthropic/.../rawPredict), not the generic OpenAI-compatible MaaS endpoint used by
+// xAI and open models. Mistral has a separate native rawPredict branch below.
 function isClaudeModel(model: string) {
-  return model.toLowerCase().startsWith("claude-");
+  return getVertexModelTransport(model) === "anthropic";
+}
+
+function isMistralModel(model: string) {
+  return getVertexModelTransport(model) === "mistral";
+}
+
+interface VertexUrlCredentials {
+  apiKey?: string | null;
+  accessToken?: string | null;
+  projectId?: string;
+  providerSpecificData?: {
+    projectId?: string;
+    project?: string;
+    region?: string;
+  };
+}
+
+function configuredVertexProjectId(credentials?: VertexUrlCredentials | null): string | undefined {
+  return (
+    credentials?.projectId ||
+    credentials?.providerSpecificData?.projectId ||
+    credentials?.providerSpecificData?.project
+  );
+}
+
+function projectIdFromServiceAccount(
+  credentials?: VertexUrlCredentials | null
+): string | undefined {
+  if (!credentials?.apiKey) return undefined;
+  try {
+    const sa = parseSAFromApiKey(credentials.apiKey);
+    if (sa.project_id) return sa.project_id;
+  } catch {
+    // Ignored, handled in execute
+  }
+  return undefined;
+}
+
+function encodeOpaqueApiKey(credentials?: VertexUrlCredentials | null): string | null {
+  if (!isExpressApiKey(credentials?.apiKey) || credentials?.accessToken) return null;
+  return encodeURIComponent(String(credentials.apiKey).trim());
+}
+
+function buildExpressGeminiUrl(
+  canonicalModel: string,
+  stream: boolean,
+  expressKey: string
+): string {
+  if (getVertexModelTransport(canonicalModel) !== "gemini") {
+    throw new Error(
+      "Vertex partner models require project-scoped credentials; Express API keys support Gemini models only"
+    );
+  }
+  const op = stream ? "streamGenerateContent?alt=sse&" : "generateContent?";
+  return `https://aiplatform.googleapis.com/v1/publishers/google/models/${canonicalModel}:${op}key=${expressKey}`;
+}
+
+function buildProjectScopedVertexUrl(
+  canonicalModel: string,
+  stream: boolean,
+  project: string,
+  region: string,
+  opaqueApiKey: string | null
+): string {
+  const apiKeySuffix = opaqueApiKey ? `?key=${opaqueApiKey}` : "";
+  if (isClaudeModel(canonicalModel)) {
+    // streamRawPredict?alt=sse was verified to return a single plain JSON body (not real SSE
+    // framing) rather than actual chunked events, which breaks the SSE parser upstream
+    // ("stream ended before producing a non-ping SSE event"). rawPredict is confirmed reliable
+    // for both streaming and non-streaming requests; always use it here.
+    return `https://aiplatform.googleapis.com/v1/projects/${project}/locations/${region}/publishers/anthropic/models/${canonicalModel}:rawPredict${apiKeySuffix}`;
+  }
+  if (isMistralModel(canonicalModel)) {
+    const operation = stream ? "streamRawPredict" : "rawPredict";
+    return `https://aiplatform.googleapis.com/v1/projects/${project}/locations/${region}/publishers/mistralai/models/${canonicalModel}:${operation}${apiKeySuffix}`;
+  }
+  if (isPartnerModel(canonicalModel)) {
+    return `https://aiplatform.googleapis.com/v1/projects/${project}/locations/global/endpoints/openapi/chat/completions${apiKeySuffix}`;
+  }
+  const operation = stream ? "streamGenerateContent?alt=sse" : "generateContent";
+  const querySeparator = opaqueApiKey ? (stream ? "&" : "?") : "";
+  return `https://aiplatform.googleapis.com/v1/projects/${project}/locations/${region}/publishers/google/models/${canonicalModel}:${operation}${querySeparator}${opaqueApiKey ? `key=${opaqueApiKey}` : ""}`;
 }
 
 // Defensive normalizer: target-format resolution for manually-added custom Claude models under
@@ -153,7 +225,8 @@ function isClaudeModel(model: string) {
 // converts a Gemini-shaped body to Anthropic Messages shape so the executor works either way,
 // independent of that unresolved upstream resolution gap.
 function toAnthropicBody(body: Record<string, unknown>): Record<string, unknown> {
-  const contents = body.contents as Array<{ role?: string; parts?: Array<{ text?: string }> }> | undefined;
+  const contents = body.contents as
+    Array<{ role?: string; parts?: Array<{ text?: string }> }> | undefined;
   if (!Array.isArray(contents)) return body;
 
   const messages = contents.map((c) => ({
@@ -161,7 +234,8 @@ function toAnthropicBody(body: Record<string, unknown>): Record<string, unknown>
     content: (c.parts || []).map((p) => p.text || "").join(""),
   }));
   const generationConfig = body.generationConfig as { maxOutputTokens?: number } | undefined;
-  const systemInstruction = body.systemInstruction as { parts?: Array<{ text?: string }> } | undefined;
+  const systemInstruction = body.systemInstruction as
+    { parts?: Array<{ text?: string }> } | undefined;
 
   const converted: Record<string, unknown> = {
     messages,
@@ -244,7 +318,11 @@ function synthesizeClaudeSse(response: Record<string, unknown>): string {
     } else if (block.type === "thinking") {
       events.push({
         event: "content_block_start",
-        data: { type: "content_block_start", index, content_block: { type: "thinking", thinking: "" } },
+        data: {
+          type: "content_block_start",
+          index,
+          content_block: { type: "thinking", thinking: "" },
+        },
       });
       if (block.thinking) {
         events.push({
@@ -280,6 +358,21 @@ export class VertexExecutor extends BaseExecutor {
   }
 
   async execute(input: ExecuteInput) {
+    const canonicalModel = normalizeVertexModelId(input.model);
+    let canonicalBody = input.body;
+    if (canonicalBody && typeof canonicalBody === "object" && !Array.isArray(canonicalBody)) {
+      const bodyRecord = canonicalBody as Record<string, unknown>;
+      if (typeof bodyRecord.model === "string") {
+        const canonicalBodyModel = normalizeVertexModelId(bodyRecord.model);
+        if (canonicalBodyModel !== bodyRecord.model) {
+          canonicalBody = { ...bodyRecord, model: canonicalBodyModel };
+        }
+      }
+    }
+    if (canonicalModel !== input.model || canonicalBody !== input.body) {
+      input = { ...input, model: canonicalModel, body: canonicalBody };
+    }
+
     const { credentials, log, model, stream } = input;
     // Defensive: trim stray surrounding whitespace from a pasted credential.
     if (typeof credentials.apiKey === "string") {
@@ -287,12 +380,17 @@ export class VertexExecutor extends BaseExecutor {
     }
     // Service Account JSON → mint a short-lived OAuth token (Bearer). An Express-mode API key is
     // sent as-is via x-goog-api-key (see buildHeaders), so no token exchange is needed for it.
-    if (credentials.apiKey && !credentials.accessToken && looksLikeServiceAccountJson(credentials.apiKey)) {
+    if (
+      credentials.apiKey &&
+      !credentials.accessToken &&
+      looksLikeServiceAccountJson(credentials.apiKey)
+    ) {
       try {
         const sa = parseSAFromApiKey(credentials.apiKey);
         credentials.accessToken = await getAccessToken(sa);
-      } catch (err: any) {
-        log?.error?.("VERTEX", `Failed to generate JWT token: ${err.message}`);
+      } catch (err) {
+        const message = err instanceof Error ? err.message : String(err);
+        log?.error?.("VERTEX", `Failed to generate JWT token: ${message}`);
         throw err;
       }
     }
@@ -318,7 +416,10 @@ export class VertexExecutor extends BaseExecutor {
       const response = result instanceof Response ? result : result?.response;
       if (response?.ok) {
         const contentType = response.headers.get("content-type") || "";
-        if (contentType.includes("application/json") && !contentType.includes("text/event-stream")) {
+        if (
+          contentType.includes("application/json") &&
+          !contentType.includes("text/event-stream")
+        ) {
           const jsonText = await response.text();
           let newBody = jsonText;
           let newContentType = contentType;
@@ -345,46 +446,28 @@ export class VertexExecutor extends BaseExecutor {
     return result;
   }
 
-  buildUrl(model: string, stream: boolean, urlIndex = 0, credentials: any = null) {
+  buildUrl(
+    model: string,
+    stream: boolean,
+    _urlIndex = 0,
+    credentials: VertexUrlCredentials | null = null
+  ) {
+    const canonicalModel = normalizeVertexModelId(model);
+    const configuredProject = configuredVertexProjectId(credentials);
+    const opaqueApiKey = encodeOpaqueApiKey(credentials);
     // Vertex AI Express mode: project-less v1 publisher endpoint with the API key passed as a
-    // ?key= query parameter (verified working contract — same as the CaptionAI GeminiClient). The
-    // Express key is NOT accepted as a Bearer/OAuth credential or via x-goog-api-key on this API.
-    if (isExpressApiKey(credentials?.apiKey) && !credentials?.accessToken) {
-      const expressKey = encodeURIComponent(String(credentials.apiKey).trim());
-      if (isPartnerModel(model)) {
-        // Partner (Anthropic/etc.) models are not available via Express keys; best-effort.
-        return `https://aiplatform.googleapis.com/v1/publishers/openapi/chat/completions?key=${expressKey}`;
-      }
-      const op = stream ? "streamGenerateContent?alt=sse&" : "generateContent?";
-      return `https://aiplatform.googleapis.com/v1/publishers/google/models/${model}:${op}key=${expressKey}`;
+    // ?key= query parameter. Express currently exposes Gemini models only; opaque Authorization
+    // Keys can use project-scoped APIs when a projectId is configured on the connection.
+    if (opaqueApiKey && !configuredProject) {
+      return buildExpressGeminiUrl(canonicalModel, stream, opaqueApiKey);
     }
-
+    const project =
+      projectIdFromServiceAccount(credentials) || configuredProject || "unknown-project";
     const region = credentials?.providerSpecificData?.region || "us-central1";
-    let project = "unknown-project";
-
-    if (credentials?.apiKey) {
-      try {
-        const sa = parseSAFromApiKey(credentials.apiKey);
-        if (sa.project_id) project = sa.project_id;
-      } catch {
-        // Ignored, handled in execute
-      }
-    }
-
-    if (isClaudeModel(model)) {
-      // streamRawPredict?alt=sse was verified to return a single plain JSON body (not real SSE
-      // framing) rather than actual chunked events, which breaks the SSE parser upstream
-      // ("stream ended before producing a non-ping SSE event"). rawPredict is confirmed reliable
-      // for both streaming and non-streaming requests; always use it here.
-      return `https://aiplatform.googleapis.com/v1/projects/${project}/locations/${region}/publishers/anthropic/models/${model}:rawPredict`;
-    }
-    if (isPartnerModel(model)) {
-      return `https://aiplatform.googleapis.com/v1/projects/${project}/locations/global/endpoints/openapi/chat/completions`;
-    }
-    return `https://aiplatform.googleapis.com/v1/projects/${project}/locations/${region}/publishers/google/models/${model}:${stream ? "streamGenerateContent?alt=sse" : "generateContent"}`;
+    return buildProjectScopedVertexUrl(canonicalModel, stream, project, region, opaqueApiKey);
   }
 
-  buildHeaders(credentials: any, stream = true) {
+  buildHeaders(credentials: VertexUrlCredentials, stream = true) {
     const headers: Record<string, string> = { "Content-Type": "application/json" };
     if (credentials.accessToken) {
       headers["Authorization"] = `Bearer ${credentials.accessToken}`;
diff --git a/src/app/(dashboard)/dashboard/providers/[id]/__tests__/useModelImportHandlers.test.tsx b/src/app/(dashboard)/dashboard/providers/[id]/__tests__/useModelImportHandlers.test.tsx
index f9909c92f95..f6c4c056472 100644
--- a/src/app/(dashboard)/dashboard/providers/[id]/__tests__/useModelImportHandlers.test.tsx
+++ b/src/app/(dashboard)/dashboard/providers/[id]/__tests__/useModelImportHandlers.test.tsx
@@ -294,3 +294,47 @@ describe("useModelImportHandlers — upstream model auto-fetch", () => {
     expect(fetchConnections).toHaveBeenCalled();
   });
 });
+
+describe("useModelImportHandlers — imported model metadata", () => {
+  it("preserves a discovered per-model targetFormat", async () => {
+    const hook = renderHook(
+      buildParams({
+        providerId: "vertex",
+        connections: [conn("vertex-connection", true)],
+      })
+    );
+    const fetchMock = vi.mocked(fetch);
+    fetchMock
+      .mockResolvedValueOnce({
+        ok: true,
+        json: async () => ({
+          models: [
+            {
+              id: "grok-4.6",
+              name: "Grok 4.6",
+              apiFormat: "chat-completions",
+              supportedEndpoints: ["chat"],
+              targetFormat: "openai",
+            },
+          ],
+        }),
+      } as Response)
+      .mockResolvedValue({ ok: true } as Response);
+
+    await act(async () => {
+      await hook.get().handleImportModels();
+    });
+
+    const createCall = fetchMock.mock.calls.find(
+      ([url, init]) => url === "/api/provider-models" && init?.method === "POST"
+    );
+    expect(createCall).toBeDefined();
+    expect(JSON.parse(String(createCall?.[1]?.body))).toEqual(
+      expect.objectContaining({
+        provider: "vertex",
+        modelId: "grok-4.6",
+        targetFormat: "openai",
+      })
+    );
+  });
+});
diff --git a/src/app/(dashboard)/dashboard/providers/[id]/components/modals/AddApiKeyModal.tsx b/src/app/(dashboard)/dashboard/providers/[id]/components/modals/AddApiKeyModal.tsx
index 3611433feb1..9fd9439a658 100644
--- a/src/app/(dashboard)/dashboard/providers/[id]/components/modals/AddApiKeyModal.tsx
+++ b/src/app/(dashboard)/dashboard/providers/[id]/components/modals/AddApiKeyModal.tsx
@@ -184,13 +184,15 @@ export default function AddApiKeyModal({
     ? providerText(t, "modalTokenIdLabel", "Token ID")
     : isAwsPolly
       ? providerText(t, "awsPollySecretAccessKeyLabel", "AWS Secret Access Key")
-      : isQoder
-        ? t("personalAccessTokenLabel")
-        : webSessionCredential
-          ? getWebSessionCredentialLabel(t, webSessionCredential, apiKeyOptional)
-          : apiKeyOptional
-            ? `${t("apiKeyLabel")} (${t("optional").toLowerCase()})`
-            : t("apiKeyLabel");
+      : isVertex
+        ? providerText(t, "vertexCredentialLabel", "API Key or Service Account JSON")
+        : isQoder
+          ? t("personalAccessTokenLabel")
+          : webSessionCredential
+            ? getWebSessionCredentialLabel(t, webSessionCredential, apiKeyOptional)
+            : apiKeyOptional
+              ? `${t("apiKeyLabel")} (${t("optional").toLowerCase()})`
+              : t("apiKeyLabel");
   const apiCredentialPlaceholder = isModal
     ? "ak-xxxxxxxxxxxxxxxx"
     : isVertex
@@ -210,19 +212,25 @@ export default function AddApiKeyModal({
         "modalTokenIdHint",
         "Modal auth uses a Token ID + Token Secret pair. Create one at https://modal.com/settings → API Tokens."
       )
-    : isQoder
-      ? t("qoderPatHint")
-      : isFreebuff
-        ? "Freebuff uses an authentic CLI auth token obtained via codebuff CLI login or automated harvester."
-        : isWebSessionCredential
-          ? getWebSessionCredentialHint(t, webSessionCredential, providerDisplayName, false)
-          : isLocalSelfHostedProvider
-            ? t("localProviderApiKeyOptionalHint", {
-                provider: localProviderMetadata?.name || providerName || provider || "",
-              })
-            : apiKeyOptional
-              ? t("apiKeyOptionalHint")
-              : undefined;
+    : isVertex
+      ? providerText(
+          t,
+          "vertexCredentialHint",
+          "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
+        )
+      : isQoder
+        ? t("qoderPatHint")
+        : isFreebuff
+          ? "Freebuff uses an authentic CLI auth token obtained via codebuff CLI login or automated harvester."
+          : isWebSessionCredential
+            ? getWebSessionCredentialHint(t, webSessionCredential, providerDisplayName, false)
+            : isLocalSelfHostedProvider
+              ? t("localProviderApiKeyOptionalHint", {
+                  provider: localProviderMetadata?.name || providerName || provider || "",
+                })
+              : apiKeyOptional
+                ? t("apiKeyOptionalHint")
+                : undefined;
   const credentialValidationFailedMessage = isWebSessionCredential
     ? providerText(
         t,
diff --git a/src/app/(dashboard)/dashboard/providers/[id]/components/modals/EditConnectionModal.tsx b/src/app/(dashboard)/dashboard/providers/[id]/components/modals/EditConnectionModal.tsx
index 59d7328fbda..347c5ec3e54 100644
--- a/src/app/(dashboard)/dashboard/providers/[id]/components/modals/EditConnectionModal.tsx
+++ b/src/app/(dashboard)/dashboard/providers/[id]/components/modals/EditConnectionModal.tsx
@@ -57,6 +57,7 @@ import { assignEditApiKeyProviderSpecificData } from "./connectionProviderSpecif
 import { isM365TierCapableProvider, normalizeM365TierValue, type M365TierValue } from "./m365Tier";
 import ProviderTierField from "./ProviderTierField";
 import AgentrouterConsoleFields from "./AgentrouterConsoleFields";
+import { getVertexCredentialCopy } from "./vertexCredentialCopy";
 import QuotaScrapingFields, { EMPTY_QUOTA_SCRAPING_FIELDS } from "./QuotaScrapingFields";
 import GlmTeamQuotaFields, { EMPTY_GLM_TEAM_QUOTA_FIELDS } from "./GlmTeamQuotaFields";
 import ProviderRegionField, { getProviderRegionConfig } from "./AlibabaProviderRegionField";
@@ -245,27 +246,25 @@ export default function EditConnectionModal({
   const isCcCompatible = isClaudeCodeCompatibleProvider(provider);
   const isCompatible =
     isOpenAICompatibleProvider(provider) || isAnthropicCompatibleProvider(provider);
+  const vertexCopy = isVertex ? getVertexCredentialCopy(t) : null;
   const apiCredentialLabel = webSessionCredential
     ? getWebSessionCredentialLabel(t, webSessionCredential, apiKeyOptional)
     : isAwsPolly
       ? providerText(t, "awsPollySecretAccessKeyLabel", "AWS Secret Access Key")
-      : apiKeyOptional
-        ? t("apiKeyOptionalLabel")
-        : t("apiKeyLabel");
+      : (vertexCopy?.label ?? (apiKeyOptional ? t("apiKeyOptionalLabel") : t("apiKeyLabel")));
   const apiCredentialPlaceholder = isWebSessionCredential
     ? webSessionCredential.placeholder
-    : isVertex
-      ? t("vertexServiceAccountPlaceholder")
-      : t("enterNewApiKey");
+    : (vertexCopy?.placeholder ?? t("enterNewApiKey"));
   const apiCredentialHint = isWebSessionCredential
     ? getWebSessionCredentialHint(t, webSessionCredential, providerDisplayName, true)
-    : isLocalSelfHostedProvider
-      ? t("localProviderApiKeyOptionalHint", {
-          provider: localProviderMetadata?.name || provider || "",
-        })
-      : apiKeyOptional
-        ? t("apiKeyOptionalHint")
-        : t("leaveBlankKeepCurrentApiKey");
+    : (vertexCopy?.hint ??
+      (isLocalSelfHostedProvider
+        ? t("localProviderApiKeyOptionalHint", {
+            provider: localProviderMetadata?.name || provider || "",
+          })
+        : apiKeyOptional
+          ? t("apiKeyOptionalHint")
+          : t("leaveBlankKeepCurrentApiKey")));
   // Modal-open form initialization from the loaded connection — applied as a
   // render-phase adjustment guarded by the previously initialized connection
   // (react.dev "adjusting state when a prop changes") instead of the former
diff --git a/src/app/(dashboard)/dashboard/providers/[id]/components/modals/vertexCredentialCopy.ts b/src/app/(dashboard)/dashboard/providers/[id]/components/modals/vertexCredentialCopy.ts
new file mode 100644
index 00000000000..70af8e72b35
--- /dev/null
+++ b/src/app/(dashboard)/dashboard/providers/[id]/components/modals/vertexCredentialCopy.ts
@@ -0,0 +1,14 @@
+import { providerText, type ProviderMessageTranslator } from "../../providerPageHelpers";
+
+/** Vertex-specific credential label/placeholder/hint for the connection modals. */
+export function getVertexCredentialCopy(t: ProviderMessageTranslator) {
+  return {
+    label: providerText(t, "vertexCredentialLabel", "API Key or Service Account JSON"),
+    placeholder: t("vertexServiceAccountPlaceholder"),
+    hint: providerText(
+      t,
+      "vertexCredentialHint",
+      "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
+    ),
+  };
+}
diff --git a/src/app/(dashboard)/dashboard/providers/[id]/hooks/useModelImportHandlers.ts b/src/app/(dashboard)/dashboard/providers/[id]/hooks/useModelImportHandlers.ts
index 99370e606e0..88121833bcd 100644
--- a/src/app/(dashboard)/dashboard/providers/[id]/hooks/useModelImportHandlers.ts
+++ b/src/app/(dashboard)/dashboard/providers/[id]/hooks/useModelImportHandlers.ts
@@ -230,6 +230,7 @@ export function useModelImportHandlers({
             ...(Array.isArray(model.supportedEndpoints)
               ? { supportedEndpoints: model.supportedEndpoints }
               : {}),
+            ...(typeof model.targetFormat === "string" ? { targetFormat: model.targetFormat } : {}),
           }),
         });
         if (!modelAliases[baseAlias]) {
diff --git a/src/app/api/providers/[id]/models/route.ts b/src/app/api/providers/[id]/models/route.ts
index c09a2636b7d..da35a319459 100755
--- a/src/app/api/providers/[id]/models/route.ts
+++ b/src/app/api/providers/[id]/models/route.ts
@@ -87,7 +87,6 @@ import {
 } from "@/lib/providerModels/modelDiscovery";
 import { buildProviderModelsUrl, getDiscoveryClientVersionOptions } from "./discoveryClientVersion";
 import { getAdobeModels } from "./adobeFireflyDiscovery";
-import { parseGeminiModelsList } from "@/lib/providerModels/geminiModelsParser";
 import { getSyncedAvailableModels, getCustomModels, getModelIsHidden } from "@/lib/db/models";
 import { isConnectionUnavailableToAuxiliaryActivity } from "@/lib/exclusiveLeaseIsolation";
 import { fetchCursorAgentModels } from "@/lib/providerModels/cursorAgent";
@@ -129,6 +128,7 @@ import {
   fetchCodexGithubCatalogModels,
 } from "./discovery/codex";
 import { maybeHandleConolModelDiscovery } from "./conolDiscovery";
+import { maybeHandleVertexModelDiscovery } from "./vertexDiscovery";
 import { buildNoAuthModelsResponse, filterModelsForRoute } from "./modelRouteProjection";
 
 /**
@@ -616,9 +616,7 @@ export async function GET(
       try {
         const discovery = await discoverMaxaiModels({
           providerSpecificData: connection.providerSpecificData as
-            | Record<string, unknown>
-            | null
-            | undefined,
+            Record<string, unknown> | null | undefined,
           accessToken: apiKey || accessToken,
           fetchImpl: (url, init) =>
             safeOutboundFetch(url, {
@@ -1788,169 +1786,22 @@ export async function GET(
       });
     }
 
-    if (provider === "vertex" || provider === "vertex-partner") {
-      const cachedResponse = maybeReturnCachedDiscovery();
-      if (cachedResponse) return cachedResponse;
-
-      const autoFetchDisabledResponse = maybeReturnAutoFetchDisabled();
-      if (autoFetchDisabledResponse) return autoFetchDisabledResponse;
-
-      // Vertex AI lists models from the Generative Language `v1beta/models` endpoint, which both
-      // Express-mode API keys (via ?key=) and Service Account JSON (via a minted OAuth Bearer
-      // token) can reach. This surfaces the live catalog, including gemini-*-image models
-      // absent from the static registry list.
-      const credential = (apiKey || "").trim();
-      let queryKey: string | null = null;
-      let bearerToken: string | null = null;
-      try {
-        const { parseSAFromApiKey, getAccessToken } =
-          await import("@omniroute/open-sse/executors/vertex.ts");
-        if (accessToken) {
-          bearerToken = accessToken;
-        } else if (credential) {
-          // A Service Account credential is a JSON object; a Vertex AI Express-mode API key is an
-          // opaque (non-JSON) string. Detect locally so this branch has no dependency on optional
-          // executor helpers.
-          let isServiceAccountJson = false;
-          try {
-            const parsed = JSON.parse(credential);
-            isServiceAccountJson = !!parsed && typeof parsed === "object" && !Array.isArray(parsed);
-          } catch {
-            isServiceAccountJson = false;
-          }
-
-          if (isServiceAccountJson) {
-            bearerToken = await getAccessToken(parseSAFromApiKey(credential));
-          } else {
-            queryKey = credential;
-          }
-        }
-      } catch (error) {
-        // Couldn't resolve a usable credential (e.g. malformed Service Account JSON).
-        const fallback = buildDiscoveryErrorFallbackResponse(error, {
-          cacheWarning: "Vertex credential unavailable — using cached catalog",
-          localWarning: "Vertex credential unavailable — using local catalog",
-        });
-        if (fallback) return fallback;
-      }
-
-      if (!queryKey && !bearerToken) {
-        const fallback = buildDiscoveryFallbackResponse({
-          cacheWarning: "No usable Vertex credential — using cached catalog",
-          localWarning: "No usable Vertex credential — using local catalog",
-        });
-        if (fallback) return fallback;
-        return NextResponse.json(
-          { error: "No usable Vertex AI credential configured for model discovery." },
-          { status: 400 }
-        );
-      }
-
-      const baseUrl = "https://generativelanguage.googleapis.com/v1beta/models?pageSize=1000";
-      const headers: Record<string, string> = { "Content-Type": "application/json" };
-      if (bearerToken) headers["Authorization"] = `Bearer ${bearerToken}`;
-
-      const allModels: any[] = [];
-      let pageUrl = queryKey ? `${baseUrl}&key=${encodeURIComponent(queryKey)}` : baseUrl;
-      let pageCount = 0;
-      const MAX_PAGES = 20;
-      const seenTokens = new Set<string>();
-
-      try {
-        while (pageUrl && pageCount < MAX_PAGES) {
-          pageCount++;
-          const response = await safeOutboundFetch(pageUrl, {
-            ...SAFE_OUTBOUND_FETCH_PRESETS.modelsPagination,
-            guard: getProviderOutboundGuard(),
-            proxyConfig: proxy,
-            method: "GET",
-            headers,
-          });
-
-          if (!response.ok) {
-            // Avoid logging the raw upstream body (may contain sensitive data); status is enough.
-            console.log("[models] Vertex model discovery failed", {
-              provider,
-              status: response.status,
-            });
-            const fallback = buildDiscoveryFallbackResponse();
-            if (fallback) return fallback;
-            return NextResponse.json(
-              { error: `Failed to fetch Vertex models: ${response.status}` },
-              { status: response.status }
-            );
-          }
-
-          const data = await response.json();
-          allModels.push(...parseGeminiModelsList(data));
-
-          const nextPageToken = data.nextPageToken;
-          if (!nextPageToken || seenTokens.has(nextPageToken)) break;
-          seenTokens.add(nextPageToken);
-          pageUrl = `${baseUrl}&pageToken=${encodeURIComponent(nextPageToken)}`;
-          if (queryKey) pageUrl += `&key=${encodeURIComponent(queryKey)}`;
-        }
-      } catch (error) {
-        const fallback = buildDiscoveryErrorFallbackResponse(error);
-        if (fallback) return fallback;
-        throw error;
-      }
-
-      // Anthropic partner models via Model Garden publisher endpoint (Bearer only).
-      //
-      // Model Garden's publisher-model LIST is served by the v1beta1 API — the v1
-      // API does not support list operations (every /v1/.../publishers/anthropic/models
-      // path 404s at the Google Front End). The list is also global: it returns the
-      // full Anthropic Claude catalog regardless of the connection's project or
-      // region, so no project/region scoping is applied here (execution region is
-      // handled separately by the vertex executor at request time).
-      if (bearerToken) {
-        const anthropicModelsUrl =
-          "https://aiplatform.googleapis.com/v1beta1/publishers/anthropic/models";
-
-        try {
-          const anthropicResponse = await safeOutboundFetch(anthropicModelsUrl, {
-            ...SAFE_OUTBOUND_FETCH_PRESETS.modelsDiscovery,
-            guard: getProviderOutboundGuard(),
-            proxyConfig: proxy,
-            method: "GET",
-            headers: {
-              "Content-Type": "application/json",
-              Authorization: `Bearer ${bearerToken}`,
-            },
-          });
-          if (anthropicResponse.ok) {
-            const anthropicData = await anthropicResponse.json();
-            const { parseVertexAnthropicModels } =
-              await import("@/lib/providerModels/vertexAnthropicModelsParser");
-            allModels.push(...parseVertexAnthropicModels(anthropicData));
-          } else {
-            console.log("[models] Vertex Anthropic partner discovery failed", {
-              provider,
-              status: anthropicResponse.status,
-            });
-          }
-        } catch (err) {
-          console.log("[models] Vertex Anthropic partner discovery error", {
-            provider,
-            error: err instanceof Error ? err.message : String(err),
-          });
-        }
-      }
-
-      if (allModels.length > 0) {
-        return buildApiDiscoveryResponse(allModels);
-      }
-
-      const fallback = buildDiscoveryFallbackResponse();
-      if (fallback) return fallback;
-      return buildResponse({
-        provider,
-        connectionId,
-        models: [],
-        source: "api",
-      });
-    }
+    const vertexResponse = await maybeHandleVertexModelDiscovery({
+      provider,
+      connectionId,
+      connection,
+      apiKey,
+      accessToken,
+      proxy,
+      cachedDiscoveryModels,
+      maybeReturnCachedDiscovery,
+      maybeReturnAutoFetchDisabled,
+      buildDiscoveryFallbackResponse,
+      buildDiscoveryErrorFallbackResponse,
+      buildResponse,
+      buildApiDiscoveryResponse,
+    });
+    if (vertexResponse) return vertexResponse;
 
     if (isAnthropicCompatibleProvider(provider)) {
       // CC providers never support models listing — this check must precede
diff --git a/src/app/api/providers/[id]/models/vertexDiscovery.ts b/src/app/api/providers/[id]/models/vertexDiscovery.ts
new file mode 100644
index 00000000000..c1173668182
--- /dev/null
+++ b/src/app/api/providers/[id]/models/vertexDiscovery.ts
@@ -0,0 +1,260 @@
+import { NextResponse } from "next/server";
+import { getModelsByProviderId } from "@/shared/constants/models";
+import { getStaticModelsForProvider } from "@/lib/providers/staticModels";
+import { updateProviderConnection } from "@/lib/db/providers";
+import { SAFE_OUTBOUND_FETCH_PRESETS, safeOutboundFetch } from "@/shared/network/safeOutboundFetch";
+import { getProviderOutboundGuard } from "@/shared/network/outboundUrlGuardPolicy";
+import type { VertexMetadataModel } from "@/lib/providerModels/vertexModelMetadata";
+import { asRecord, mergeLocalCatalogModels, toNonEmptyString } from "./discovery/helpers";
+
+interface DiscoveryWarnings {
+  cacheWarning?: string;
+  localWarning?: string;
+}
+
+interface VertexDiscoveryRouteOptions {
+  provider: string;
+  connectionId: string;
+  connection: { projectId?: unknown; providerSpecificData?: unknown };
+  apiKey: unknown;
+  accessToken: unknown;
+  proxy: unknown;
+  cachedDiscoveryModels: VertexMetadataModel[];
+  maybeReturnCachedDiscovery: () => Response | null;
+  maybeReturnAutoFetchDisabled: () => Response | null;
+  buildDiscoveryFallbackResponse: (warnings?: DiscoveryWarnings) => Response | null;
+  buildDiscoveryErrorFallbackResponse: (
+    error: unknown,
+    warnings?: DiscoveryWarnings
+  ) => Response | null;
+  buildResponse: (payload: Record<string, unknown>) => Response;
+  buildApiDiscoveryResponse: (
+    models: unknown[],
+    warning?: string,
+    extraPayload?: Record<string, unknown>
+  ) => Promise<Response>;
+}
+
+type CuratedVertexModel = { id: string; name: string; owned_by: string };
+
+function asNamedModels(models: unknown[]): Array<{ id: string; name?: string }> {
+  return models.flatMap((model) =>
+    model && typeof model === "object" && "id" in model && typeof model.id === "string"
+      ? [model as { id: string; name?: string }]
+      : []
+  );
+}
+
+function vertexFetch(proxy: unknown): (url: string, init: RequestInit) => Promise<Response> {
+  return (url, init) =>
+    safeOutboundFetch(url, {
+      ...SAFE_OUTBOUND_FETCH_PRESETS.modelsPagination,
+      guard: getProviderOutboundGuard(),
+      proxyConfig: proxy,
+      ...init,
+    });
+}
+
+function vertexMetadataFetch(
+  proxy: unknown
+): (url: string, init: RequestInit) => Promise<Response> {
+  return (url, init) =>
+    safeOutboundFetch(url, {
+      ...SAFE_OUTBOUND_FETCH_PRESETS.modelsDiscovery,
+      guard: "public-only",
+      proxyConfig: proxy,
+      ...init,
+    });
+}
+
+function curatedVertexCatalog(provider: string): CuratedVertexModel[] {
+  return mergeLocalCatalogModels(
+    getModelsByProviderId(provider) || [],
+    getStaticModelsForProvider(provider) || []
+  ).map((model) => ({
+    ...model,
+    name: model.name || model.id,
+    owned_by: provider,
+  }));
+}
+
+async function resolveVertexDiscoveryAuth(options: {
+  apiKey: unknown;
+  accessToken: unknown;
+}): Promise<{ queryKey: string | null; bearerToken: string | null }> {
+  const { parseSAFromApiKey, getAccessToken, looksLikeServiceAccountJson } =
+    await import("@omniroute/open-sse/executors/vertex.ts");
+  const credential = (typeof options.apiKey === "string" ? options.apiKey : "").trim();
+  if (typeof options.accessToken === "string" && options.accessToken) {
+    return { queryKey: null, bearerToken: options.accessToken };
+  }
+  if (!credential) return { queryKey: null, bearerToken: null };
+  if (looksLikeServiceAccountJson(credential)) {
+    return {
+      queryKey: null,
+      bearerToken: await getAccessToken(parseSAFromApiKey(credential)),
+    };
+  }
+  return { queryKey: credential, bearerToken: null };
+}
+
+async function handleVertexApiKeyCatalog(
+  options: VertexDiscoveryRouteOptions,
+  queryKey: string,
+  catalog: CuratedVertexModel[]
+): Promise<Response> {
+  const { discoverVertexModelsWithApiKey } =
+    await import("@/lib/providerModels/vertexModelDiscovery");
+  const discovery = await discoverVertexModelsWithApiKey({
+    apiKey: queryKey,
+    fetchImpl: vertexFetch(options.proxy),
+  });
+
+  if (discovery.models.length === 0 && discovery.unavailable) {
+    const warning = "Vertex model discovery temporarily unavailable — using local catalog";
+    return (
+      options.buildDiscoveryFallbackResponse({
+        cacheWarning: "Vertex model discovery temporarily unavailable — using cached catalog",
+        localWarning: warning,
+      }) ??
+      options.buildResponse({
+        provider: options.provider,
+        connectionId: options.connectionId,
+        models: [],
+        source: "local_catalog",
+        warning,
+      })
+    );
+  }
+
+  const providerData = asRecord(options.connection.providerSpecificData);
+  const configuredProjectId =
+    toNonEmptyString(options.connection.projectId) ||
+    toNonEmptyString(providerData.projectId) ||
+    toNonEmptyString(providerData.project);
+  const projectId = configuredProjectId || discovery.projectId || null;
+  const projectIdAutoDetected = !configuredProjectId && !!discovery.projectId;
+  if (projectIdAutoDetected && projectId) {
+    await updateProviderConnection(options.connectionId, { projectId });
+  }
+
+  const { isVertexExpressModel } = await import("@omniroute/open-sse/config/vertexModels.ts");
+  if (projectId) {
+    const liveGeminiModels = asNamedModels(discovery.models);
+    const projectCatalog = mergeLocalCatalogModels(liveGeminiModels, catalog);
+    if (liveGeminiModels.length > 0) {
+      return options.buildApiDiscoveryResponse(projectCatalog, undefined, {
+        projectIdAutoDetected,
+        catalogMode: "live_gemini_curated_project",
+      });
+    }
+    return options.buildResponse({
+      provider: options.provider,
+      connectionId: options.connectionId,
+      models: projectCatalog,
+      source: "local_catalog",
+      intentional: true,
+      projectIdAutoDetected,
+      catalogMode: "curated_project",
+    });
+  }
+
+  if (discovery.models.length > 0) {
+    return options.buildApiDiscoveryResponse(discovery.models, discovery.warning);
+  }
+
+  return options.buildResponse({
+    provider: options.provider,
+    connectionId: options.connectionId,
+    models: catalog.filter((model) => isVertexExpressModel(model.id)),
+    source: "local_catalog",
+    intentional: true,
+    warning:
+      (discovery.failureStatus
+        ? `Generative Language model listing rejected the API key (HTTP ${discovery.failureStatus}). `
+        : "") + "No live catalog available for this API key — using curated Express catalog",
+  });
+}
+
+export async function maybeHandleVertexModelDiscovery(
+  options: VertexDiscoveryRouteOptions
+): Promise<Response | null> {
+  if (options.provider !== "vertex" && options.provider !== "vertex-partner") return null;
+
+  const cachedResponse = options.maybeReturnCachedDiscovery();
+  if (cachedResponse) return cachedResponse;
+
+  const autoFetchDisabledResponse = options.maybeReturnAutoFetchDisabled();
+  if (autoFetchDisabledResponse) return autoFetchDisabledResponse;
+
+  let queryKey: string | null = null;
+  let bearerToken: string | null = null;
+  try {
+    const resolved = await resolveVertexDiscoveryAuth({
+      apiKey: options.apiKey,
+      accessToken: options.accessToken,
+    });
+    queryKey = resolved.queryKey;
+    bearerToken = resolved.bearerToken;
+  } catch (error) {
+    const fallback = options.buildDiscoveryErrorFallbackResponse(error, {
+      cacheWarning: "Vertex credential unavailable — using cached catalog",
+      localWarning: "Vertex credential unavailable — using local catalog",
+    });
+    if (fallback) return fallback;
+  }
+
+  if (!queryKey && !bearerToken) {
+    const fallback = options.buildDiscoveryFallbackResponse({
+      cacheWarning: "No usable Vertex credential — using cached catalog",
+      localWarning: "No usable Vertex credential — using local catalog",
+    });
+    if (fallback) return fallback;
+    return NextResponse.json(
+      { error: "No usable Vertex AI credential configured for model discovery." },
+      { status: 400 }
+    );
+  }
+
+  const catalog = curatedVertexCatalog(options.provider);
+  if (queryKey) return handleVertexApiKeyCatalog(options, queryKey, catalog);
+
+  const { discoverVertexModelsWithBearer } =
+    await import("@/lib/providerModels/vertexModelDiscovery");
+  const discovery = await discoverVertexModelsWithBearer({
+    bearerToken: bearerToken as string,
+    fetchImpl: vertexFetch(options.proxy),
+  });
+
+  if (discovery.models.length > 0) {
+    const liveModels = asNamedModels(discovery.models);
+    let enrichedLiveModels = liveModels;
+    try {
+      const { enrichVertexModelsWithMetadata } =
+        await import("@/lib/providerModels/vertexModelMetadata");
+      enrichedLiveModels = await enrichVertexModelsWithMetadata({
+        models: liveModels,
+        staleModels: options.cachedDiscoveryModels,
+        fetchImpl: vertexMetadataFetch(options.proxy),
+      });
+    } catch {
+      // Metadata is optional. A documentation or parser failure must never turn an authenticated
+      // live Vertex catalog into a failed model sync.
+    }
+    return options.buildApiDiscoveryResponse(enrichedLiveModels, discovery.warning, {
+      catalogMode: "live_vertex_catalog",
+    });
+  }
+
+  const fallback = options.buildDiscoveryFallbackResponse({
+    cacheWarning: "Vertex model catalogs unavailable — using cached catalog",
+    localWarning: "Vertex model catalogs unavailable — using local catalog",
+  });
+  if (fallback) return fallback;
+  return options.buildResponse({
+    provider: options.provider,
+    connectionId: options.connectionId,
+    models: [],
+    source: "api",
+  });
+}
diff --git a/src/app/api/v1beta/models/route.ts b/src/app/api/v1beta/models/route.ts
index bf9ddec175d..d9233b4f433 100644
--- a/src/app/api/v1beta/models/route.ts
+++ b/src/app/api/v1beta/models/route.ts
@@ -99,7 +99,12 @@ export async function GET() {
           displayName: m.name || m.id,
           ...(typeof m.description === "string" ? { description: m.description } : {}),
           supportedGenerationMethods: ["generateContent"],
-          inputTokenLimit: typeof m.inputTokenLimit === "number" ? m.inputTokenLimit : 128000,
+          inputTokenLimit:
+            typeof m.inputTokenLimit === "number"
+              ? m.inputTokenLimit
+              : typeof m.contextWindow === "number"
+                ? m.contextWindow
+                : 128000,
           outputTokenLimit: typeof m.outputTokenLimit === "number" ? m.outputTokenLimit : 8192,
           ...(m.supportsThinking === true ? { thinking: true } : {}),
         });
@@ -131,7 +136,9 @@ export async function GET() {
             inputTokenLimit:
               typeof m.inputTokenLimit === "number"
                 ? m.inputTokenLimit
-                : resolved.maxInputTokens || resolved.contextWindow || 128000,
+                : typeof m.contextWindow === "number"
+                  ? m.contextWindow
+                  : resolved.maxInputTokens || resolved.contextWindow || 128000,
             outputTokenLimit:
               typeof m.outputTokenLimit === "number"
                 ? m.outputTokenLimit
@@ -173,7 +180,9 @@ export async function GET() {
             inputTokenLimit:
               typeof m.inputTokenLimit === "number"
                 ? m.inputTokenLimit
-                : resolved.maxInputTokens || resolved.contextWindow || 128000,
+                : typeof m.contextWindow === "number"
+                  ? m.contextWindow
+                  : resolved.maxInputTokens || resolved.contextWindow || 128000,
             outputTokenLimit:
               typeof m.outputTokenLimit === "number"
                 ? m.outputTokenLimit
diff --git a/src/i18n/messages/ar.json b/src/i18n/messages/ar.json
index 1d7e80a0c06..fc632b54871 100644
--- a/src/i18n/messages/ar.json
+++ b/src/i18n/messages/ar.json
@@ -5963,6 +5963,8 @@
     "validationModelIdHint": "النموذج المستخدم للتحقق من الاتصال",
     "validationModelIdLabel": "معرّف نموذج التحقق",
     "validationModelIdPlaceholder": "مثال: gpt-4o-mini",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery.",
     "vertexServiceAccountPlaceholder": "ألصق JSON لحساب الخدمة ({\"type\":\"service_account\",\"project_id\":\"…\",\"client_email\":\"…\",\"private_key\":\"…\"}) أو OAuth access_token",
     "webCookieProviders": "مزودو ملفات تعريف ارتباط الويب",
     "weeklyShort": "أسبوعي",
diff --git a/src/i18n/messages/az.json b/src/i18n/messages/az.json
index f6ca25914a7..128cc27d288 100644
--- a/src/i18n/messages/az.json
+++ b/src/i18n/messages/az.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Saxlanmış model son nöqtəsi parametrləri",
     "searchByModelAria": "Model üzrə axtarış edin",
     "selectSupportedEndpoint": "Ən azı bir dəstəklənən son nöqtəni seçin",
-    "antigravityClientProfileHarness": "Test sistemi / CLI"
+    "antigravityClientProfileHarness": "Test sistemi / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Parametrlər",
diff --git a/src/i18n/messages/bg.json b/src/i18n/messages/bg.json
index 5ea8ef1e4c8..ebca6eaeab6 100644
--- a/src/i18n/messages/bg.json
+++ b/src/i18n/messages/bg.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Настройки на крайна точка на запазен модел",
     "searchByModelAria": "Търсене по модел",
     "selectSupportedEndpoint": "Изберете поне една поддържана крайна точка",
-    "antigravityClientProfileHarness": "Тестова среда / CLI"
+    "antigravityClientProfileHarness": "Тестова среда / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Настройки",
diff --git a/src/i18n/messages/bn.json b/src/i18n/messages/bn.json
index 33d59055656..509b86ed1c4 100644
--- a/src/i18n/messages/bn.json
+++ b/src/i18n/messages/bn.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "সংরক্ষিত মডেল এন্ডপয়েন্ট সেটিংস",
     "searchByModelAria": "মডেল দ্বারা অনুসন্ধান করুন",
     "selectSupportedEndpoint": "কমপক্ষে একটি সমর্থিত এন্ডপয়েন্ট নির্বাচন করুন",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "সেটিংস",
diff --git a/src/i18n/messages/cs.json b/src/i18n/messages/cs.json
index a6d823ccdc0..ddf8bdcbb8b 100644
--- a/src/i18n/messages/cs.json
+++ b/src/i18n/messages/cs.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Nastavení koncového bodu uloženého modelu",
     "searchByModelAria": "Hledat podle modelu",
     "selectSupportedEndpoint": "Vyberte alespoň jeden podporovaný koncový bod",
-    "antigravityClientProfileHarness": "Testovací rozhraní / CLI"
+    "antigravityClientProfileHarness": "Testovací rozhraní / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Nastavení",
diff --git a/src/i18n/messages/da.json b/src/i18n/messages/da.json
index f345293d50f..9004e921e13 100644
--- a/src/i18n/messages/da.json
+++ b/src/i18n/messages/da.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Indstillinger for gemt model endpoint",
     "searchByModelAria": "Søg efter model",
     "selectSupportedEndpoint": "Vælg mindst én understøttet endpoint",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Indstillinger",
diff --git a/src/i18n/messages/de.json b/src/i18n/messages/de.json
index e4d6ac4ce5c..a3bbf4ca96e 100644
--- a/src/i18n/messages/de.json
+++ b/src/i18n/messages/de.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Einstellungen für den gespeicherten Modell-Endpunkt",
     "searchByModelAria": "Nach Modell suchen",
     "selectSupportedEndpoint": "Wählen Sie mindestens einen unterstützten Endpunkt aus",
-    "antigravityClientProfileHarness": "Testsystem / CLI"
+    "antigravityClientProfileHarness": "Testsystem / CLI",
+    "vertexCredentialLabel": "API-Key oder Service-Account-JSON",
+    "vertexCredentialHint": "API-Keys verwenden den kuratierten Projektkatalog. Ein Service-Account-JSON aktiviert die Live-Erkennung im Model Garden."
   },
   "settings": {
     "title": "Einstellungen",
diff --git a/src/i18n/messages/en.json b/src/i18n/messages/en.json
index 0a685beed5a..e8cb19eb44a 100644
--- a/src/i18n/messages/en.json
+++ b/src/i18n/messages/en.json
@@ -5963,7 +5963,9 @@
     "validationModelIdHint": "Model used to verify the API key. Leave blank to use the provider’s first available model.",
     "validationModelIdLabel": "Validation Model",
     "validationModelIdPlaceholder": "e.g. meta-llama/llama-3.1-8b-instruct",
-    "vertexServiceAccountPlaceholder": "Paste your Service Account JSON ({\"type\":\"service_account\",\"project_id\":\"…\",\"client_email\":\"…\",\"private_key\":\"…\"}) or an OAuth access_token",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery.",
+    "vertexServiceAccountPlaceholder": "Paste an API key, Service Account JSON ({\"type\":\"service_account\",\"project_id\":\"…\",\"client_email\":\"…\",\"private_key\":\"…\"}), or OAuth access token",
     "webCookieProviders": "Web Cookie Providers",
     "weeklyShort": "Weekly Short",
     "xiaomiMimoBaseUrlHint": "Xiaomi Mimo Base Url Hint",
diff --git a/src/i18n/messages/es.json b/src/i18n/messages/es.json
index 5588aabd008..a234d4afaf7 100644
--- a/src/i18n/messages/es.json
+++ b/src/i18n/messages/es.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Configuración del punto final del modelo guardado",
     "searchByModelAria": "Buscar por modelo",
     "selectSupportedEndpoint": "Seleccione al menos un endpoint compatible",
-    "antigravityClientProfileHarness": "Entorno / CLI"
+    "antigravityClientProfileHarness": "Entorno / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Configuración",
diff --git a/src/i18n/messages/fa.json b/src/i18n/messages/fa.json
index d8742159bd6..e4f54135888 100644
--- a/src/i18n/messages/fa.json
+++ b/src/i18n/messages/fa.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "تنظیمات نقطه پایانی مدل ذخیره شده",
     "searchByModelAria": "جستجو بر اساس مدل",
     "selectSupportedEndpoint": "حداقل یک نقطه پایانی پشتیبانی شده را انتخاب کنید",
-    "antigravityClientProfileHarness": "هارنس / CLI"
+    "antigravityClientProfileHarness": "هارنس / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "تنظیمات",
diff --git a/src/i18n/messages/fi.json b/src/i18n/messages/fi.json
index b5b5c054da8..003b0b62a30 100644
--- a/src/i18n/messages/fi.json
+++ b/src/i18n/messages/fi.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Tallennetun mallin päätepisteen asetukset",
     "searchByModelAria": "Hae mallin mukaan",
     "selectSupportedEndpoint": "Valitse vähintään yksi tuettu päätepiste",
-    "antigravityClientProfileHarness": "Testauskehys / CLI"
+    "antigravityClientProfileHarness": "Testauskehys / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Asetukset",
diff --git a/src/i18n/messages/fr.json b/src/i18n/messages/fr.json
index af56b66d8c5..d48a1b7f194 100644
--- a/src/i18n/messages/fr.json
+++ b/src/i18n/messages/fr.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Saved modèles endpoint paramètres",
     "searchByModelAria": "Rechercher un modèle",
     "selectSupportedEndpoint": "Sélectionnez au moins un endpoint pris en charge",
-    "antigravityClientProfileHarness": "Harnais / CLI"
+    "antigravityClientProfileHarness": "Harnais / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Paramètres",
diff --git a/src/i18n/messages/gu.json b/src/i18n/messages/gu.json
index 92fc0f0ede9..734ab7194d2 100644
--- a/src/i18n/messages/gu.json
+++ b/src/i18n/messages/gu.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "સાચવેલ મોડેલ અંતિમ બિંદુની સેટિંગ્સ",
     "searchByModelAria": "મોડલ દ્વારા શોધો",
     "selectSupportedEndpoint": "કમથી કમ એક સમર્થિત અંતિમ બિંદુ પસંદ કરો",
-    "antigravityClientProfileHarness": "હાર્નેસ / CLI"
+    "antigravityClientProfileHarness": "હાર્નેસ / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "સેટિંગ્સ",
diff --git a/src/i18n/messages/he.json b/src/i18n/messages/he.json
index 034fbd6991a..586097d48a8 100644
--- a/src/i18n/messages/he.json
+++ b/src/i18n/messages/he.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "הגדרות נקודת הקצה של המודל השמור",
     "searchByModelAria": "חפש לפי דגם",
     "selectSupportedEndpoint": "בחר לפחות נקודת קצה אחת נתמכת",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "הגדרות",
diff --git a/src/i18n/messages/hi.json b/src/i18n/messages/hi.json
index 473f9f45e86..d385465c89f 100644
--- a/src/i18n/messages/hi.json
+++ b/src/i18n/messages/hi.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "सहेजे गए मॉडल एंडपॉइंट सेटिंग्स",
     "searchByModelAria": "मॉडल द्वारा खोजें",
     "selectSupportedEndpoint": "कम से कम एक समर्थित एंडपॉइंट चुनें",
-    "antigravityClientProfileHarness": "हार्नेस / CLI"
+    "antigravityClientProfileHarness": "हार्नेस / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "सेटिंग्स",
diff --git a/src/i18n/messages/hu.json b/src/i18n/messages/hu.json
index c8943186f16..2d586907916 100644
--- a/src/i18n/messages/hu.json
+++ b/src/i18n/messages/hu.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Mentett modell végpont beállításai",
     "searchByModelAria": "Keresés modell szerint",
     "selectSupportedEndpoint": "Válasszon ki legalább egy támogatott végpontot",
-    "antigravityClientProfileHarness": "Tesztkeretrendszer / CLI"
+    "antigravityClientProfileHarness": "Tesztkeretrendszer / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Beállítások elemre",
diff --git a/src/i18n/messages/id.json b/src/i18n/messages/id.json
index 5976b326491..2c7bba2540f 100644
--- a/src/i18n/messages/id.json
+++ b/src/i18n/messages/id.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Pengaturan endpoint model yang disimpan",
     "searchByModelAria": "Cari berdasarkan model",
     "selectSupportedEndpoint": "Pilih setidaknya satu endpoint yang didukung",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Pengaturan",
diff --git a/src/i18n/messages/it.json b/src/i18n/messages/it.json
index 39dfe2aeef8..a6985019012 100644
--- a/src/i18n/messages/it.json
+++ b/src/i18n/messages/it.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Impostazioni dell'endpoint del modello salvato",
     "searchByModelAria": "Cerca per modello",
     "selectSupportedEndpoint": "Seleziona almeno un endpoint supportato",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Impostazioni",
diff --git a/src/i18n/messages/ja.json b/src/i18n/messages/ja.json
index edad2e2f6b2..b7620f50673 100644
--- a/src/i18n/messages/ja.json
+++ b/src/i18n/messages/ja.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "保存されたモデルエンドポイント設定",
     "searchByModelAria": "モデルで検索",
     "selectSupportedEndpoint": "サポートされているエンドポイントを少なくとも1つ選択してください",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "設定",
diff --git a/src/i18n/messages/ko.json b/src/i18n/messages/ko.json
index 6d4047278f1..c3c3ec7f510 100644
--- a/src/i18n/messages/ko.json
+++ b/src/i18n/messages/ko.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "저장된 모델 엔드포인트 설정",
     "searchByModelAria": "모델로 검색",
     "selectSupportedEndpoint": "지원되는 엔드포인트를 최소한 하나 선택하세요.",
-    "antigravityClientProfileHarness": "하네스 / CLI"
+    "antigravityClientProfileHarness": "하네스 / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "설정",
diff --git a/src/i18n/messages/mr.json b/src/i18n/messages/mr.json
index 036a6215c69..1dedfc2e32a 100644
--- a/src/i18n/messages/mr.json
+++ b/src/i18n/messages/mr.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "सुरक्षित केलेल्या मॉडेल एंडपॉइंट सेटिंग्ज",
     "searchByModelAria": "मॉडेलद्वारे शोधा",
     "selectSupportedEndpoint": "किमान एक समर्थित एंडपॉइंट निवडा",
-    "antigravityClientProfileHarness": "हार्नेस / CLI"
+    "antigravityClientProfileHarness": "हार्नेस / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "सेटिंग्ज",
diff --git a/src/i18n/messages/ms.json b/src/i18n/messages/ms.json
index 82d40caada4..1524c32ca8c 100644
--- a/src/i18n/messages/ms.json
+++ b/src/i18n/messages/ms.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Tetapan titik akhir model yang disimpan",
     "searchByModelAria": "Cari mengikut model",
     "selectSupportedEndpoint": "Pilih sekurang-kurangnya satu titik akhir yang disokong",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "tetapan",
diff --git a/src/i18n/messages/nl.json b/src/i18n/messages/nl.json
index 95eb77fb447..4dfd626472e 100644
--- a/src/i18n/messages/nl.json
+++ b/src/i18n/messages/nl.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Instellingen voor opgeslagen model-eindpunt",
     "searchByModelAria": "Zoeken op model",
     "selectSupportedEndpoint": "Selecteer ten minste één ondersteunde eindpunt",
-    "antigravityClientProfileHarness": "Testomgeving / CLI"
+    "antigravityClientProfileHarness": "Testomgeving / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Instellingen",
diff --git a/src/i18n/messages/no.json b/src/i18n/messages/no.json
index 1b561286bf3..6f099cd72a3 100644
--- a/src/i18n/messages/no.json
+++ b/src/i18n/messages/no.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Innstillinger for lagrede modellendepunkter",
     "searchByModelAria": "Søk etter modell",
     "selectSupportedEndpoint": "Velg minst ett støttet endepunkt",
-    "antigravityClientProfileHarness": "Testverktøy / CLI"
+    "antigravityClientProfileHarness": "Testverktøy / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Innstillinger",
diff --git a/src/i18n/messages/phi.json b/src/i18n/messages/phi.json
index 4f7ea295922..4e70b90ac73 100644
--- a/src/i18n/messages/phi.json
+++ b/src/i18n/messages/phi.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Naka-save na mga setting ng endpoint ng modelo",
     "searchByModelAria": "Maghanap ayon sa modelo",
     "selectSupportedEndpoint": "Pumili ng hindi bababa sa isang sinusuportahang endpoint",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Mga setting",
diff --git a/src/i18n/messages/pl.json b/src/i18n/messages/pl.json
index 454cbb4cd2e..f2a72f1523d 100644
--- a/src/i18n/messages/pl.json
+++ b/src/i18n/messages/pl.json
@@ -5963,6 +5963,8 @@
     "validationModelIdHint": "Model używany do weryfikacji klucza API. Pozostaw puste, aby użyć pierwszego dostępnego model dla provider.",
     "validationModelIdLabel": "Model weryfikacyjny",
     "validationModelIdPlaceholder": "np. meta-llama/llama-3.1-8b-instruct",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery.",
     "vertexServiceAccountPlaceholder": "Wklej Service Account JSON ({\"type\":\"service_account\",\"project_id\":\"…\",\"client_email\":\"…\",\"private_key\":\"…\"}) lub OAuth access_token",
     "webCookieProviders": "Dostawcy korzystający z ciasteczek WWW",
     "weeklyShort": "Tygodniowo",
diff --git a/src/i18n/messages/pt-BR.json b/src/i18n/messages/pt-BR.json
index 5d83758096a..8f79ebef64d 100644
--- a/src/i18n/messages/pt-BR.json
+++ b/src/i18n/messages/pt-BR.json
@@ -5963,6 +5963,8 @@
     "validationModelIdHint": "Usado como fallback se a listagem de modelos não estiver disponível.",
     "validationModelIdLabel": "ID do Modelo (opcional)",
     "validationModelIdPlaceholder": "ex.: grok-3 ou meta-llama/Llama-3.1-8B-Instruct",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery.",
     "vertexServiceAccountPlaceholder": "Cole o JSON da Service Account aqui",
     "webCookieProviders": "Provedores Web / Cookie",
     "weeklyShort": "Semanal",
diff --git a/src/i18n/messages/pt.json b/src/i18n/messages/pt.json
index 136e9920861..9c70c9333c8 100644
--- a/src/i18n/messages/pt.json
+++ b/src/i18n/messages/pt.json
@@ -5963,6 +5963,8 @@
     "validationModelIdHint": "Validation Model Id Hint",
     "validationModelIdLabel": "Validation Model Id Label",
     "validationModelIdPlaceholder": "Validation Model Id Placeholder",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery.",
     "vertexServiceAccountPlaceholder": "Cola o teu JSON de Service Account ({\"type\":\"service_account\",\"project_id\":\"…\",\"client_email\":\"…\",\"private_key\":\"…\"})",
     "webCookieProviders": "Fornecedores de Cookies Web",
     "weeklyShort": "Semanal",
diff --git a/src/i18n/messages/ro.json b/src/i18n/messages/ro.json
index 9c103d9ba8e..933d501d2ff 100644
--- a/src/i18n/messages/ro.json
+++ b/src/i18n/messages/ro.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Setările punctului final al modelului salvat",
     "searchByModelAria": "Caută după model",
     "selectSupportedEndpoint": "Selectați cel puțin un punct final acceptat",
-    "antigravityClientProfileHarness": "Cadru / CLI"
+    "antigravityClientProfileHarness": "Cadru / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Setări",
diff --git a/src/i18n/messages/ru.json b/src/i18n/messages/ru.json
index 697938d8dfb..b7f8b341539 100644
--- a/src/i18n/messages/ru.json
+++ b/src/i18n/messages/ru.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Настройки конечной точки сохраненной модели",
     "searchByModelAria": "Поиск по модели",
     "selectSupportedEndpoint": "Выберите хотя бы одну поддерживаемую конечную точку",
-    "antigravityClientProfileHarness": "Среда / CLI"
+    "antigravityClientProfileHarness": "Среда / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Настройки",
diff --git a/src/i18n/messages/sk.json b/src/i18n/messages/sk.json
index 7e1a6d557af..c611a1e736d 100644
--- a/src/i18n/messages/sk.json
+++ b/src/i18n/messages/sk.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Nastavenia koncového bodu uloženého modelu",
     "searchByModelAria": "Hľadať podľa modelu",
     "selectSupportedEndpoint": "Vyberte aspoň jeden podporovaný koncový bod",
-    "antigravityClientProfileHarness": "Testovacie prostredie / CLI"
+    "antigravityClientProfileHarness": "Testovacie prostredie / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Nastavenia",
diff --git a/src/i18n/messages/sv.json b/src/i18n/messages/sv.json
index 5b0bfe83500..f98d3aadeb1 100644
--- a/src/i18n/messages/sv.json
+++ b/src/i18n/messages/sv.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Inställningar för sparad modellslutpunkt",
     "searchByModelAria": "Sök efter modell",
     "selectSupportedEndpoint": "Välj minst en stödd slutpunkt",
-    "antigravityClientProfileHarness": "Testmiljö / CLI"
+    "antigravityClientProfileHarness": "Testmiljö / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Inställningar",
diff --git a/src/i18n/messages/sw.json b/src/i18n/messages/sw.json
index b3f6e4fb7c0..80de72824ea 100644
--- a/src/i18n/messages/sw.json
+++ b/src/i18n/messages/sw.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Mipangilio ya mwisho wa mfano uliohifadhiwa",
     "searchByModelAria": "Tafuta kwa mfano",
     "selectSupportedEndpoint": "Chagua angalau kiunganishi kimoja kinachoungwa mkono",
-    "antigravityClientProfileHarness": "Kiunzi / CLI"
+    "antigravityClientProfileHarness": "Kiunzi / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Mipangilio",
diff --git a/src/i18n/messages/ta.json b/src/i18n/messages/ta.json
index b4de89c2a46..c501b9174d0 100644
--- a/src/i18n/messages/ta.json
+++ b/src/i18n/messages/ta.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "சேமிக்கப்பட்ட மாதிரி முடிவுறுப்பு அமைப்புகள்",
     "searchByModelAria": "மாதிரியில் தேடு",
     "selectSupportedEndpoint": "குறைந்தது ஒரு ஆதரிக்கப்படும் முடிவுகளைத் தேர்ந்தெடுக்கவும்",
-    "antigravityClientProfileHarness": "ஹார்னஸ் / CLI"
+    "antigravityClientProfileHarness": "ஹார்னஸ் / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "அமைப்புகள்",
diff --git a/src/i18n/messages/te.json b/src/i18n/messages/te.json
index ab48cf6f812..c95427b21df 100644
--- a/src/i18n/messages/te.json
+++ b/src/i18n/messages/te.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "సేవ్ చేసిన మోడల్ ఎండ్‌పాయింట్ సెట్టింగ్స్",
     "searchByModelAria": "మోడల్ ద్వారా శోధించండి",
     "selectSupportedEndpoint": "కమిషన్ చేయబడిన కనెక్ట్ చేయబడిన ఎండ్‌పాయింట్‌లలో కనీసం ఒకటి ఎంచుకోండి",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "సెట్టింగ్లు",
diff --git a/src/i18n/messages/th.json b/src/i18n/messages/th.json
index 8d512b93d43..ede97a216be 100644
--- a/src/i18n/messages/th.json
+++ b/src/i18n/messages/th.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "การตั้งค่า endpoint ของโมเดลที่บันทึกไว้",
     "searchByModelAria": "ค้นหาตามรุ่น",
     "selectSupportedEndpoint": "เลือกจุดสิ้นสุดที่รองรับอย่างน้อยหนึ่งจุด",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "การตั้งค่า",
diff --git a/src/i18n/messages/tr.json b/src/i18n/messages/tr.json
index 8eef16e5131..5ab58cb192d 100644
--- a/src/i18n/messages/tr.json
+++ b/src/i18n/messages/tr.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Kaydedilmiş model uç noktası ayarları",
     "searchByModelAria": "Model ile ara",
     "selectSupportedEndpoint": "En az bir desteklenen uç noktayı seçin",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Ayarlar",
diff --git a/src/i18n/messages/uk-UA.json b/src/i18n/messages/uk-UA.json
index 5b95cd89dde..17d64b60aef 100644
--- a/src/i18n/messages/uk-UA.json
+++ b/src/i18n/messages/uk-UA.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "Налаштування кінцевої точки збереженої моделі",
     "searchByModelAria": "Пошук за моделлю",
     "selectSupportedEndpoint": "Виберіть принаймні одну підтримувану точку доступу",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "Налаштування",
diff --git a/src/i18n/messages/ur.json b/src/i18n/messages/ur.json
index af87e321f1d..009a15ef547 100644
--- a/src/i18n/messages/ur.json
+++ b/src/i18n/messages/ur.json
@@ -6478,7 +6478,9 @@
     "savedModelEndpointSettings": "محفوظ شدہ ماڈل اینڈپوائنٹ کی ترتیبات",
     "searchByModelAria": "ماڈل کے ذریعے تلاش کریں",
     "selectSupportedEndpoint": "کم از کم ایک سپورٹ کردہ اینڈپوائنٹ منتخب کریں",
-    "antigravityClientProfileHarness": "Harness / CLI"
+    "antigravityClientProfileHarness": "Harness / CLI",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery."
   },
   "settings": {
     "title": "ترتیبات",
diff --git a/src/i18n/messages/vi.json b/src/i18n/messages/vi.json
index f1673cbc487..5908b33964f 100644
--- a/src/i18n/messages/vi.json
+++ b/src/i18n/messages/vi.json
@@ -5963,6 +5963,8 @@
     "validationModelIdHint": "Mô hình được sử dụng để xác minh khóa API. Để trống để sử dụng mô hình khả dụng đầu tiên của nhà cung cấp.",
     "validationModelIdLabel": "Mô hình xác thực",
     "validationModelIdPlaceholder": "Ví dụ: meta-llama/llama-3.1-8b-instruct",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery.",
     "vertexServiceAccountPlaceholder": "Dán JSON Tài khoản dịch vụ của bạn ({\"type\":\"service_account\",\"project_id\":\"…\",\"client_email\":\"…\",\"private_key\":\"…\"}) hoặc access_token OAuth",
     "webCookieProviders": "Nhà cung cấp Web Cookie",
     "weeklyShort": "Hàng tuần",
diff --git a/src/i18n/messages/zh-CN.json b/src/i18n/messages/zh-CN.json
index d1b93cb084b..05bd00fa510 100644
--- a/src/i18n/messages/zh-CN.json
+++ b/src/i18n/messages/zh-CN.json
@@ -5963,6 +5963,8 @@
     "validationModelIdHint": "用于验证此提供者连接的模型 ID。",
     "validationModelIdLabel": "验证模型 ID",
     "validationModelIdPlaceholder": "输入用于测试的模型 ID",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery.",
     "vertexServiceAccountPlaceholder": "粘贴 Service Account JSON（{\"type\":\"service_account\"...}）或 OAuth access_token",
     "webCookieProviders": "Web Cookie 提供者",
     "weeklyShort": "每周",
diff --git a/src/i18n/messages/zh-TW.json b/src/i18n/messages/zh-TW.json
index 4ed7fbecceb..4eee594e494 100644
--- a/src/i18n/messages/zh-TW.json
+++ b/src/i18n/messages/zh-TW.json
@@ -5963,6 +5963,8 @@
     "validationModelIdHint": "用於驗證此提供者連線的模型 ID。",
     "validationModelIdLabel": "驗證模型 ID",
     "validationModelIdPlaceholder": "輸入用於測試的模型 ID",
+    "vertexCredentialLabel": "API Key or Service Account JSON",
+    "vertexCredentialHint": "API keys use the curated project catalog. Service Account JSON enables live Model Garden discovery.",
     "vertexServiceAccountPlaceholder": "貼上 Service Account JSON（{\"type\":\"service_account\"...}）或 OAuth access_token",
     "webCookieProviders": "Web Cookie Provider",
     "weeklyShort": "每週",
diff --git a/src/lib/combos/builderOptions.ts b/src/lib/combos/builderOptions.ts
index 0f1cb529cca..945479b2c9a 100644
--- a/src/lib/combos/builderOptions.ts
+++ b/src/lib/combos/builderOptions.ts
@@ -30,6 +30,7 @@ type CustomModelLike = {
   apiFormat?: string;
   supportedEndpoints?: string[];
   inputTokenLimit?: number;
+  contextWindow?: number;
   outputTokenLimit?: number;
   supportsThinking?: boolean;
   isHidden?: boolean;
@@ -41,6 +42,7 @@ type SyncedModelLike = {
   source?: string;
   supportedEndpoints?: string[];
   inputTokenLimit?: number;
+  contextWindow?: number;
   outputTokenLimit?: number;
   description?: string;
   supportsThinking?: boolean;
@@ -384,7 +386,10 @@ function buildModelOptions(
       name: toStringOrNull(model.name),
       source: "imported",
       supportedEndpoints: toStringArray(model.supportedEndpoints),
-      contextLength: toNumberOrNull(model.inputTokenLimit) ?? resolved.contextWindow,
+      contextLength:
+        toNumberOrNull(model.contextWindow) ??
+        toNumberOrNull(model.inputTokenLimit) ??
+        resolved.contextWindow,
       outputTokenLimit: toNumberOrNull(model.outputTokenLimit) ?? resolved.maxOutputTokens,
       supportsThinking:
         typeof model.supportsThinking === "boolean"
@@ -528,7 +533,7 @@ function buildModelOptions(
       source,
       supportedEndpoints: toStringArray(model.supportedEndpoints),
       apiFormat: toStringOrNull(model.apiFormat),
-      contextLength: toNumberOrNull(model.inputTokenLimit),
+      contextLength: toNumberOrNull(model.contextWindow) ?? toNumberOrNull(model.inputTokenLimit),
       outputTokenLimit: toNumberOrNull(model.outputTokenLimit),
       supportsThinking:
         typeof model.supportsThinking === "boolean" ? model.supportsThinking : undefined,
diff --git a/src/lib/contextWindowResolver.ts b/src/lib/contextWindowResolver.ts
index 8e7124ea85f..fc8c9895ba7 100644
--- a/src/lib/contextWindowResolver.ts
+++ b/src/lib/contextWindowResolver.ts
@@ -75,12 +75,19 @@ export function reconcileContextWindows(
 
 /** Flatten the per-provider discovery map into the reconcile input. */
 function toDiscoveredWindows(
-  byProvider: Record<string, Array<{ id: string; inputTokenLimit?: number }>>
+  byProvider: Record<
+    string,
+    Array<{ id: string; contextWindow?: number; inputTokenLimit?: number }>
+  >
 ): DiscoveredWindow[] {
   const out: DiscoveredWindow[] = [];
   for (const [provider, models] of Object.entries(byProvider)) {
     for (const m of models) {
-      out.push({ provider, modelId: m.id, window: m.inputTokenLimit ?? null });
+      out.push({
+        provider,
+        modelId: m.id,
+        window: m.contextWindow ?? m.inputTokenLimit ?? null,
+      });
     }
   }
   return out;
diff --git a/src/lib/db/models/synced.ts b/src/lib/db/models/synced.ts
index 92f2c4eb7ac..1e282265a80 100644
--- a/src/lib/db/models/synced.ts
+++ b/src/lib/db/models/synced.ts
@@ -1,4 +1,5 @@
 import { isRetiredGitHubCopilotModelId } from "@omniroute/open-sse/config/providers/registry/github/retiredModels.ts";
+import type { VertexModelMetadataProvenance } from "@/lib/providerModels/vertexModelMetadata";
 
 import { asRecord, toNonEmptyString } from "./shared";
 
@@ -13,7 +14,9 @@ export interface SyncedAvailableModel {
   supportedThinkingEfforts?: string[];
   defaultThinkingEffort?: string;
   inputTokenLimit?: number;
+  contextWindow?: number;
   outputTokenLimit?: number;
+  metadataProvenance?: VertexModelMetadataProvenance;
   description?: string;
   supportsThinking?: boolean;
   alwaysThinking?: boolean;
@@ -76,9 +79,15 @@ function normalizeSyncedAvailableModel(model: unknown): SyncedAvailableModel | n
     ...(typeof record.inputTokenLimit === "number"
       ? { inputTokenLimit: record.inputTokenLimit }
       : {}),
+    ...(typeof record.contextWindow === "number" ? { contextWindow: record.contextWindow } : {}),
     ...(typeof record.outputTokenLimit === "number"
       ? { outputTokenLimit: record.outputTokenLimit }
       : {}),
+    ...(record.metadataProvenance && typeof record.metadataProvenance === "object"
+      ? {
+          metadataProvenance: record.metadataProvenance as VertexModelMetadataProvenance,
+        }
+      : {}),
     ...(typeof record.description === "string" ? { description: record.description } : {}),
     ...(typeof record.supportsThinking === "boolean"
       ? { supportsThinking: record.supportsThinking }
diff --git a/src/lib/providerModels/managedModelImport.ts b/src/lib/providerModels/managedModelImport.ts
index 8faff3613d7..173ae910710 100644
--- a/src/lib/providerModels/managedModelImport.ts
+++ b/src/lib/providerModels/managedModelImport.ts
@@ -26,6 +26,7 @@ import { isDiscoverableAgyModelId } from "@omniroute/open-sse/config/agyModels.t
 import { filterChatSelectableModels } from "@omniroute/open-sse/services/modelEndpointPolicy.ts";
 import { filterSelectableModels } from "@omniroute/open-sse/services/modelLifecycle.ts";
 import { isSelfHostedChatProvider } from "@/shared/constants/providers";
+import type { VertexModelMetadataProvenance } from "@/lib/providerModels/vertexModelMetadata";
 
 type JsonRecord = Record<string, unknown>;
 
@@ -42,7 +43,9 @@ export type ManagedImportedModel = {
   supportedThinkingEfforts?: string[];
   defaultThinkingEffort?: string;
   inputTokenLimit?: number;
+  contextWindow?: number;
   outputTokenLimit?: number;
+  metadataProvenance?: VertexModelMetadataProvenance;
   description?: string;
   supportsThinking?: boolean;
   alwaysThinking?: boolean;
@@ -77,9 +80,13 @@ function copyImportedModelMetadata(target: ManagedImportedModel, model: JsonReco
     target.defaultThinkingEffort = model.defaultThinkingEffort as string;
   }
   if (typeof model.inputTokenLimit === "number") target.inputTokenLimit = model.inputTokenLimit;
+  if (typeof model.contextWindow === "number") target.contextWindow = model.contextWindow;
   if (typeof model.outputTokenLimit === "number") {
     target.outputTokenLimit = model.outputTokenLimit;
   }
+  if (model.metadataProvenance && typeof model.metadataProvenance === "object") {
+    target.metadataProvenance = model.metadataProvenance as VertexModelMetadataProvenance;
+  }
   if (typeof model.description === "string") target.description = model.description;
   if (typeof model.supportsThinking === "boolean") {
     target.supportsThinking = model.supportsThinking;
@@ -124,6 +131,7 @@ function copyComparableModelMetadata(target: JsonRecord, model: JsonRecord): voi
     target.defaultThinkingEffort = model.defaultThinkingEffort;
   }
   if (typeof model.inputTokenLimit === "number") target.inputTokenLimit = model.inputTokenLimit;
+  if (typeof model.contextWindow === "number") target.contextWindow = model.contextWindow;
   if (typeof model.outputTokenLimit === "number") {
     target.outputTokenLimit = model.outputTokenLimit;
   }
@@ -307,7 +315,9 @@ export async function importManagedModels({
       supportedThinkingEfforts?: string[];
       defaultThinkingEffort?: string;
       inputTokenLimit?: number;
+      contextWindow?: number;
       outputTokenLimit?: number;
+      metadataProvenance?: VertexModelMetadataProvenance;
       description?: string;
       supportsThinking?: boolean;
       alwaysThinking?: boolean;
diff --git a/src/lib/providerModels/modelDiscovery.ts b/src/lib/providerModels/modelDiscovery.ts
index 9a5697c54a6..4e5cf1c3005 100644
--- a/src/lib/providerModels/modelDiscovery.ts
+++ b/src/lib/providerModels/modelDiscovery.ts
@@ -306,16 +306,20 @@ export function normalizeDiscoveredModels(
 
     const topProvider = asRecord(record.top_provider);
 
-    // OpenRouter (and similar passthrough catalogs) report the context window as
-    // `context_length` / `top_provider.context_length`, not `inputTokenLimit`.
-    // Fall back across those names so synced models carry a real window instead
-    // of the provider default (128K). Explicit `inputTokenLimit` still wins. #3202
-    const inputTokenLimit = firstPositiveNumber(
-      record.inputTokenLimit,
+    // Keep the total context window distinct from an explicit maximum-input limit. Existing
+    // providers historically stored context_length as inputTokenLimit, so retain that compatibility
+    // outside Vertex while persisting the separate contextWindow field for new consumers.
+    const contextWindow = firstPositiveNumber(
       record.context_length,
       record.contextLength,
+      record.contextWindow,
       topProvider.context_length
     );
+    const isVertexProvider = providerId === "vertex" || providerId === "vertex-partner";
+    const inputTokenLimit = firstPositiveNumber(
+      record.inputTokenLimit,
+      ...(isVertexProvider ? [] : [contextWindow])
+    );
     const outputTokenLimit = firstPositiveNumber(
       record.outputTokenLimit,
       topProvider.max_completion_tokens
@@ -346,7 +350,11 @@ export function normalizeDiscoveredModels(
       ...(supportedThinkingEfforts !== undefined ? { supportedThinkingEfforts } : {}),
       ...(defaultThinkingEffort !== undefined ? { defaultThinkingEffort } : {}),
       ...(typeof inputTokenLimit === "number" ? { inputTokenLimit } : {}),
+      ...(isVertexProvider && typeof contextWindow === "number" ? { contextWindow } : {}),
       ...(typeof outputTokenLimit === "number" ? { outputTokenLimit } : {}),
+      ...(record.metadataProvenance && typeof record.metadataProvenance === "object"
+        ? { metadataProvenance: record.metadataProvenance }
+        : {}),
       ...(typeof record.description === "string" ? { description: record.description } : {}),
       ...(typeof record.supportsThinking === "boolean"
         ? { supportsThinking: record.supportsThinking }
diff --git a/src/lib/providerModels/vertexModelDiscovery.ts b/src/lib/providerModels/vertexModelDiscovery.ts
new file mode 100644
index 00000000000..a7d99fac8ad
--- /dev/null
+++ b/src/lib/providerModels/vertexModelDiscovery.ts
@@ -0,0 +1,239 @@
+import { parseGeminiModelsList } from "@/lib/providerModels/geminiModelsParser";
+import { parseVertexPublisherModels } from "@/lib/providerModels/vertexPublisherModelsParser";
+
+const GOOGLE_MODELS_URL = "https://generativelanguage.googleapis.com/v1beta/models?pageSize=1000";
+const VERTEX_PUBLISHER_PAGE_SIZE = 300;
+const MAX_CATALOG_PAGES = 20;
+
+/**
+ * Serverless chat publishers currently documented by Vertex Model Garden. The parser and executor
+ * remain publisher-generic, so newly returned model versions need no source change.
+ */
+export const VERTEX_MODEL_GARDEN_PUBLISHERS = [
+  "google",
+  "anthropic",
+  "xai",
+  "mistralai",
+  "meta",
+  "deepseek-ai",
+  "qwen",
+  "moonshotai",
+  "minimaxai",
+  "openai",
+  "zai-org",
+] as const;
+
+export type VertexModelDiscoveryFetch = (url: string, init: RequestInit) => Promise<Response>;
+
+export interface VertexModelDiscoveryResult {
+  models: unknown[];
+  warning?: string;
+  projectId?: string;
+  failureStatus?: number;
+  unavailable?: boolean;
+}
+
+interface DiscoveryAuth {
+  headers: Record<string, string>;
+}
+
+function asObject(value: unknown): Record<string, unknown> | null {
+  return value && typeof value === "object" && !Array.isArray(value)
+    ? (value as Record<string, unknown>)
+    : null;
+}
+
+function mergeDiscoveryModelsById(models: unknown[]): unknown[] {
+  const merged = new Map<string, Record<string, unknown>>();
+  const unkeyed: unknown[] = [];
+
+  for (const candidate of models) {
+    const model = asObject(candidate);
+    if (!model) {
+      unkeyed.push(candidate);
+      continue;
+    }
+    const id = typeof model.id === "string" ? model.id : null;
+    if (!id) {
+      unkeyed.push(candidate);
+      continue;
+    }
+    const existing = merged.get(id);
+    // Earlier sources have higher precedence. The Generative Language API is queried before the
+    // publisher catalog, so its structured token limits survive while publisher transport fields
+    // fill gaps.
+    merged.set(id, existing ? { ...model, ...existing } : model);
+  }
+
+  return [...merged.values(), ...unkeyed];
+}
+
+function readNextPageToken(data: unknown): string | null {
+  const token = asObject(data)?.nextPageToken;
+  return typeof token === "string" && token.length > 0 ? token : null;
+}
+
+function readConsumerProjectId(consumer: unknown): string | null {
+  if (typeof consumer !== "string") return null;
+  const match = consumer.match(/^projects\/([^/]+)$/);
+  return match?.[1] ?? null;
+}
+
+function readApiKeyConsumerProjectId(data: unknown): string | null {
+  const details = asObject(asObject(data)?.error)?.details;
+  if (!Array.isArray(details)) return null;
+  for (const detail of details) {
+    const projectId = readConsumerProjectId(asObject(asObject(detail)?.metadata)?.consumer);
+    if (projectId) return projectId;
+  }
+  return null;
+}
+
+async function discoverVertexModels(options: {
+  auth: DiscoveryAuth;
+  fetchImpl: VertexModelDiscoveryFetch;
+}): Promise<VertexModelDiscoveryResult> {
+  const { auth, fetchImpl } = options;
+  const models: unknown[] = [];
+  let publisherFailureCount = 0;
+
+  try {
+    let pageUrl = GOOGLE_MODELS_URL;
+    let pageCount = 0;
+    const seenTokens = new Set<string>();
+
+    while (pageUrl && pageCount < MAX_CATALOG_PAGES) {
+      pageCount += 1;
+      const response = await fetchImpl(pageUrl, { method: "GET", headers: auth.headers });
+      if (!response.ok) {
+        break;
+      }
+
+      const data = await response.json();
+      models.push(...parseGeminiModelsList(data));
+      const nextPageToken = readNextPageToken(data);
+      if (!nextPageToken || seenTokens.has(nextPageToken)) break;
+      seenTokens.add(nextPageToken);
+      pageUrl = `${GOOGLE_MODELS_URL}&pageToken=${encodeURIComponent(nextPageToken)}`;
+    }
+  } catch {
+    // The Generative Language models API intentionally rejects Service Accounts. Publisher
+    // discovery below is independent and remains authoritative for the live partner catalog.
+  }
+
+  const publisherResults = await Promise.all(
+    VERTEX_MODEL_GARDEN_PUBLISHERS.map(async (publisher) => {
+      const publisherModels: unknown[] = [];
+      let pageUrl =
+        `https://aiplatform.googleapis.com/v1beta1/publishers/${publisher}/models` +
+        `?pageSize=${VERTEX_PUBLISHER_PAGE_SIZE}`;
+      let pageCount = 0;
+      const seenTokens = new Set<string>();
+
+      try {
+        while (pageUrl && pageCount < MAX_CATALOG_PAGES) {
+          pageCount += 1;
+          const response = await fetchImpl(pageUrl, { method: "GET", headers: auth.headers });
+          if (!response.ok) {
+            publisherFailureCount += 1;
+            break;
+          }
+
+          const data = await response.json();
+          publisherModels.push(...parseVertexPublisherModels(data, publisher));
+          const nextPageToken = readNextPageToken(data);
+          if (!nextPageToken || seenTokens.has(nextPageToken)) break;
+          seenTokens.add(nextPageToken);
+          pageUrl =
+            `https://aiplatform.googleapis.com/v1beta1/publishers/${publisher}/models` +
+            `?pageSize=${VERTEX_PUBLISHER_PAGE_SIZE}&pageToken=${encodeURIComponent(nextPageToken)}`;
+        }
+      } catch {
+        publisherFailureCount += 1;
+      }
+
+      return publisherModels;
+    })
+  );
+  for (const publisherModels of publisherResults) models.push(...publisherModels);
+
+  return {
+    models: mergeDiscoveryModelsById(models),
+    ...(publisherFailureCount > 0 && models.length > 0
+      ? { warning: "Some Vertex catalogs were unavailable — imported available models" }
+      : {}),
+  };
+}
+
+/** Discover live Gemini and Model Garden catalogs with OAuth or Service Account credentials. */
+export function discoverVertexModelsWithBearer(options: {
+  bearerToken: string;
+  fetchImpl: VertexModelDiscoveryFetch;
+}): Promise<VertexModelDiscoveryResult> {
+  return discoverVertexModels({
+    auth: {
+      headers: {
+        "Content-Type": "application/json",
+        Authorization: `Bearer ${options.bearerToken}`,
+      },
+    },
+    fetchImpl: options.fetchImpl,
+  });
+}
+
+/**
+ * Discover the Gemini catalog exposed to an API key and recover its Google Cloud consumer project
+ * from structured error metadata when API restrictions block models.list. Vertex Model Garden's
+ * publishers.models.list rejects API-key authentication even when the same key is authorized for
+ * project-scoped partner inference, so publisher discovery is deliberately Bearer-only.
+ */
+export async function discoverVertexModelsWithApiKey(options: {
+  apiKey: string;
+  fetchImpl: VertexModelDiscoveryFetch;
+}): Promise<VertexModelDiscoveryResult> {
+  const models: unknown[] = [];
+  const headers = {
+    "Content-Type": "application/json",
+    // Keep the secret out of URLs and any URL-bearing error/log path.
+    "x-goog-api-key": options.apiKey,
+  };
+  let pageUrl = GOOGLE_MODELS_URL;
+  let pageCount = 0;
+  const seenTokens = new Set<string>();
+
+  try {
+    while (pageUrl && pageCount < MAX_CATALOG_PAGES) {
+      pageCount += 1;
+      const response = await options.fetchImpl(pageUrl, { method: "GET", headers });
+      const data = await response.json().catch(() => null);
+      if (!response.ok) {
+        const projectId = readApiKeyConsumerProjectId(data);
+        return {
+          models,
+          failureStatus: response.status,
+          unavailable: ![400, 401, 403].includes(response.status),
+          ...(projectId ? { projectId } : {}),
+          ...(models.length > 0
+            ? { warning: "Some Vertex Gemini catalog pages were unavailable" }
+            : {}),
+        };
+      }
+
+      models.push(...parseGeminiModelsList(data));
+      const nextPageToken = readNextPageToken(data);
+      if (!nextPageToken || seenTokens.has(nextPageToken)) break;
+      seenTokens.add(nextPageToken);
+      pageUrl = `${GOOGLE_MODELS_URL}&pageToken=${encodeURIComponent(nextPageToken)}`;
+    }
+  } catch {
+    return {
+      models,
+      unavailable: true,
+      ...(models.length > 0
+        ? { warning: "Some Vertex Gemini catalog pages were unavailable" }
+        : {}),
+    };
+  }
+
+  return { models };
+}
diff --git a/src/lib/providerModels/vertexModelMetadata.ts b/src/lib/providerModels/vertexModelMetadata.ts
new file mode 100644
index 00000000000..53409dfb9ba
--- /dev/null
+++ b/src/lib/providerModels/vertexModelMetadata.ts
@@ -0,0 +1,379 @@
+const VERTEX_DOCS_BASE = "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models";
+const VERTEX_DOCS_PARSER_VERSION = "vertex-docs-v1";
+const FRESH_CACHE_TTL_MS = 24 * 60 * 60 * 1000;
+const STALE_CACHE_TTL_MS = 7 * 24 * 60 * 60 * 1000;
+const MAX_CONCURRENT_FETCHES = 6;
+
+export type VertexMetadataField = "contextWindow" | "inputTokenLimit" | "outputTokenLimit";
+
+export interface VertexDocsMetadataProvenance {
+  source: "google-cloud-docs";
+  sourceUrl: string;
+  fetchedAt: string;
+  lastModified?: string;
+  parserVersion: typeof VERTEX_DOCS_PARSER_VERSION;
+  confidence: "verified";
+  fields: VertexMetadataField[];
+}
+
+export interface VertexModelMetadataProvenance {
+  vertexDocs: VertexDocsMetadataProvenance;
+}
+
+export interface VertexModelMetadata {
+  contextWindow?: number;
+  inputTokenLimit?: number;
+  outputTokenLimit?: number;
+  metadataProvenance?: VertexModelMetadataProvenance;
+}
+
+export interface VertexMetadataModel {
+  id: string;
+  name?: string;
+  contextWindow?: number;
+  inputTokenLimit?: number;
+  outputTokenLimit?: number;
+  metadataProvenance?: VertexModelMetadataProvenance;
+}
+
+export type VertexMetadataFetch = (url: string, init: RequestInit) => Promise<Response>;
+
+interface ParsedDocsMetadata {
+  contextWindow?: number;
+  inputTokenLimit?: number;
+  outputTokenLimit?: number;
+}
+
+interface CachedDocsPage {
+  fetchedAtMs: number;
+  rows: string[][];
+  sourceUrl: string;
+  lastModified?: string;
+}
+
+const docsCache = new Map<string, CachedDocsPage>();
+const docsInflight = new Map<string, Promise<CachedDocsPage | null>>();
+
+function terminalModelId(modelId: string): string {
+  const segments = modelId.trim().split("/");
+  return segments[segments.length - 1] || modelId.trim();
+}
+
+function docsSlug(value: string): string {
+  return value.toLowerCase().replace(/[._]/g, "-");
+}
+
+function unique(values: string[]): string[] {
+  return Array.from(new Set(values));
+}
+
+/** Resolve official Google Cloud documentation candidates without using user-controlled hosts. */
+export function resolveVertexModelDocsUrls(modelId: string): string[] {
+  const normalized = modelId.trim().toLowerCase();
+  const bareId = terminalModelId(normalized);
+  const slug = docsSlug(bareId);
+
+  if (/^gemini-/.test(bareId)) {
+    return [`${VERTEX_DOCS_BASE}/gemini/${slug.replace(/^gemini-/, "")}?hl=en`];
+  }
+
+  if (/^claude-/.test(bareId)) {
+    return [`${VERTEX_DOCS_BASE}/partner-models/claude/${slug.replace(/^claude-/, "")}?hl=en`];
+  }
+
+  if (/^mistral-/.test(bareId)) {
+    return [`${VERTEX_DOCS_BASE}/partner-models/mistral/${slug}?hl=en`];
+  }
+
+  if (normalized.startsWith("xai/") && /^grok-/.test(bareId)) {
+    const familySlug = slug.replace(/-(?:non-)?reasoning$/, "");
+    return [`${VERTEX_DOCS_BASE}/partner-models/grok/${familySlug}?hl=en`];
+  }
+
+  const namespaceSeparator = normalized.indexOf("/");
+  if (namespaceSeparator > 0) {
+    const publisher = docsSlug(normalized.slice(0, namespaceSeparator));
+    const withoutMaasSuffix = slug.replace(/-maas$/, "");
+    return unique([
+      `${VERTEX_DOCS_BASE}/maas/${publisher}/${slug}?hl=en`,
+      `${VERTEX_DOCS_BASE}/maas/${publisher}/${withoutMaasSuffix}?hl=en`,
+    ]);
+  }
+
+  return [];
+}
+
+function parsePositiveInteger(value: string | undefined): number | undefined {
+  if (!value) return undefined;
+  const normalized = value.replace(/[\s,]/g, "");
+  if (!/^\d+$/.test(normalized)) return undefined;
+  const parsed = Number(normalized);
+  return Number.isSafeInteger(parsed) && parsed > 0 ? parsed : undefined;
+}
+
+function extractLabeledNumber(text: string, label: RegExp): number | undefined {
+  const match = text.match(new RegExp(`${label.source}\\s*:?\\s*([0-9][0-9,]*)`, "i"));
+  return parsePositiveInteger(match?.[1]);
+}
+
+function modelIdAppearsInRow(rowText: string, modelId: string): boolean {
+  const escaped = modelId.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+  return new RegExp(`(?:^|[^a-z0-9._/-])${escaped}(?:$|[^a-z0-9._/-])`, "i").test(rowText);
+}
+
+function decodeHtmlEntities(value: string): string {
+  const namedEntities: Record<string, string> = {
+    amp: "&",
+    apos: "'",
+    gt: ">",
+    lt: "<",
+    nbsp: " ",
+    quot: '"',
+  };
+
+  return value.replace(/&(#(?:x[0-9a-f]+|\d+)|[a-z]+);/gi, (entity, token: string) => {
+    if (token.startsWith("#")) {
+      const hexadecimal = token[1]?.toLowerCase() === "x";
+      const parsed = Number.parseInt(token.slice(hexadecimal ? 2 : 1), hexadecimal ? 16 : 10);
+      if (Number.isInteger(parsed) && parsed >= 0 && parsed <= 0x10ffff) {
+        return String.fromCodePoint(parsed);
+      }
+      return entity;
+    }
+    return namedEntities[token.toLowerCase()] ?? entity;
+  });
+}
+
+function htmlFragmentToText(fragment: string): string {
+  return decodeHtmlEntities(
+    fragment
+      .replace(/<(?:script|style)\b[^>]*>[\s\S]*?<\/(?:script|style)\s*>/gi, " ")
+      .replace(/<(?:br|hr)\b[^>]*\/?\s*>/gi, " ")
+      .replace(/<[^>]+>/g, " ")
+  )
+    .replace(/\s+/g, " ")
+    .trim();
+}
+
+/**
+ * Extract only the table cells needed for metadata matching. Google Cloud model pages are around
+ * 400 KB each; constructing full JSDOM windows for the live catalog expanded those pages to
+ * hundreds of megabytes. Keeping compact row text avoids both that transient heap spike and
+ * retaining raw HTML in the 24-hour cache.
+ */
+function extractTableRows(html: string): string[][] {
+  const rows: string[][] = [];
+  const rowPattern = /<tr\b[^>]*>([\s\S]*?)<\/tr\s*>/gi;
+  let rowMatch: RegExpExecArray | null;
+  while ((rowMatch = rowPattern.exec(html))) {
+    const cells: string[] = [];
+    const cellPattern = /<(th|td)\b[^>]*>([\s\S]*?)<\/\1\s*>/gi;
+    let cellMatch: RegExpExecArray | null;
+    while ((cellMatch = cellPattern.exec(rowMatch[1]))) {
+      cells.push(htmlFragmentToText(cellMatch[2]));
+    }
+    if (cells.length > 0) rows.push(cells);
+  }
+  return rows;
+}
+
+function parseVertexModelDocsRows(
+  rows: string[][],
+  expectedModelId: string
+): ParsedDocsMetadata | null {
+  const bareId = terminalModelId(expectedModelId);
+  const hasExactModelId = rows.some((cells) => {
+    const label = cells[0]?.toLowerCase();
+    const value = cells.slice(1).join(" ");
+    return label === "model id" && modelIdAppearsInRow(value, bareId);
+  });
+  if (!hasExactModelId) return null;
+
+  const parsed: ParsedDocsMetadata = {};
+  for (const cells of rows) {
+    const text = cells.join(" ");
+
+    const contextWindow =
+      extractLabeledNumber(text, /context window/i) ??
+      extractLabeledNumber(text, /context length/i);
+    const inputTokenLimit = extractLabeledNumber(text, /maximum input tokens/i);
+    const outputTokenLimit = extractLabeledNumber(text, /maximum output tokens/i);
+    if (parsed.contextWindow === undefined && contextWindow !== undefined) {
+      parsed.contextWindow = contextWindow;
+    }
+    if (parsed.inputTokenLimit === undefined && inputTokenLimit !== undefined) {
+      parsed.inputTokenLimit = inputTokenLimit;
+    }
+    if (parsed.outputTokenLimit === undefined && outputTokenLimit !== undefined) {
+      parsed.outputTokenLimit = outputTokenLimit;
+    }
+  }
+
+  return Object.keys(parsed).length > 0 ? parsed : null;
+}
+
+/**
+ * Parse token metadata only when an official semantic "Model ID" row contains the exact live id.
+ * This prevents a family, navigation, or similar-model page from lending limits to another model.
+ */
+export function parseVertexModelDocsHtml(
+  html: string,
+  expectedModelId: string
+): ParsedDocsMetadata | null {
+  return parseVertexModelDocsRows(extractTableRows(html), expectedModelId);
+}
+
+function getCachedPage(url: string, nowMs: number, maxAgeMs: number): CachedDocsPage | null {
+  const cached = docsCache.get(url);
+  return cached && nowMs - cached.fetchedAtMs <= maxAgeMs ? cached : null;
+}
+
+async function fetchDocsPage(options: {
+  url: string;
+  fetchImpl: VertexMetadataFetch;
+  nowMs: number;
+}): Promise<CachedDocsPage | null> {
+  const { url, fetchImpl, nowMs } = options;
+  const fresh = getCachedPage(url, nowMs, FRESH_CACHE_TTL_MS);
+  if (fresh) return fresh;
+
+  const existingInflight = docsInflight.get(url);
+  if (existingInflight) return existingInflight;
+
+  const request = (async () => {
+    try {
+      const response = await fetchImpl(url, {
+        method: "GET",
+        headers: { Accept: "text/html", "Accept-Language": "en" },
+      });
+      if (!response.ok) return getCachedPage(url, nowMs, STALE_CACHE_TTL_MS);
+
+      const html = await response.text();
+      const page: CachedDocsPage = {
+        fetchedAtMs: nowMs,
+        rows: extractTableRows(html),
+        sourceUrl: url,
+        ...(response.headers.get("last-modified")
+          ? { lastModified: response.headers.get("last-modified")! }
+          : {}),
+      };
+      docsCache.set(url, page);
+      return page;
+    } catch {
+      return getCachedPage(url, nowMs, STALE_CACHE_TTL_MS);
+    }
+  })().finally(() => docsInflight.delete(url));
+
+  docsInflight.set(url, request);
+  return request;
+}
+
+function metadataFields(metadata: ParsedDocsMetadata): VertexMetadataField[] {
+  return (["contextWindow", "inputTokenLimit", "outputTokenLimit"] as const).filter(
+    (field) => typeof metadata[field] === "number"
+  );
+}
+
+function readPersistedDocsMetadata(
+  model: VertexMetadataModel | undefined,
+  nowMs: number
+): VertexModelMetadata | null {
+  if (!model) return null;
+  const provenance = model.metadataProvenance as VertexModelMetadataProvenance | undefined;
+  const fetchedAtMs = Date.parse(provenance?.vertexDocs?.fetchedAt || "");
+  if (!Number.isFinite(fetchedAtMs) || nowMs - fetchedAtMs > STALE_CACHE_TTL_MS) return null;
+  const verifiedFields = new Set(provenance?.vertexDocs?.fields || []);
+
+  const metadata: VertexModelMetadata = {
+    ...(verifiedFields.has("contextWindow") && typeof model.contextWindow === "number"
+      ? { contextWindow: model.contextWindow }
+      : {}),
+    ...(verifiedFields.has("inputTokenLimit") && typeof model.inputTokenLimit === "number"
+      ? { inputTokenLimit: model.inputTokenLimit }
+      : {}),
+    ...(verifiedFields.has("outputTokenLimit") && typeof model.outputTokenLimit === "number"
+      ? { outputTokenLimit: model.outputTokenLimit }
+      : {}),
+    metadataProvenance: provenance,
+  };
+  return metadataFields(metadata).length > 0 ? metadata : null;
+}
+
+async function mapWithConcurrency<T, R>(
+  values: T[],
+  concurrency: number,
+  mapper: (value: T) => Promise<R>
+): Promise<R[]> {
+  const results = new Array<R>(values.length);
+  let nextIndex = 0;
+  const workers = Array.from({ length: Math.min(concurrency, values.length) }, async () => {
+    while (nextIndex < values.length) {
+      const index = nextIndex++;
+      results[index] = await mapper(values[index]);
+    }
+  });
+  await Promise.all(workers);
+  return results;
+}
+
+/**
+ * Enrich live Vertex models from official Google Cloud documentation. Existing structured fields
+ * win; docs fill only gaps. Failures return the live models unchanged (or verified stale values).
+ */
+export async function enrichVertexModelsWithMetadata<T extends VertexMetadataModel>(options: {
+  models: T[];
+  fetchImpl: VertexMetadataFetch;
+  staleModels?: VertexMetadataModel[];
+  now?: Date;
+}): Promise<T[]> {
+  const now = options.now || new Date();
+  const nowMs = now.getTime();
+  const staleById = new Map((options.staleModels || []).map((model) => [model.id, model]));
+
+  return mapWithConcurrency(options.models, MAX_CONCURRENT_FETCHES, async (model) => {
+    const expectedModelId = terminalModelId(model.id);
+    let docsMetadata: ParsedDocsMetadata | null = null;
+    let page: CachedDocsPage | null = null;
+
+    for (const url of resolveVertexModelDocsUrls(model.id)) {
+      page = await fetchDocsPage({
+        url,
+        fetchImpl: options.fetchImpl,
+        nowMs,
+      });
+      docsMetadata = page ? parseVertexModelDocsRows(page.rows, expectedModelId) : null;
+      if (docsMetadata) break;
+    }
+
+    const staleMetadata = readPersistedDocsMetadata(staleById.get(model.id), nowMs);
+    const metadata: VertexModelMetadata | null = docsMetadata
+      ? {
+          ...docsMetadata,
+          metadataProvenance: {
+            vertexDocs: {
+              source: "google-cloud-docs",
+              sourceUrl: page!.sourceUrl,
+              fetchedAt: new Date(page!.fetchedAtMs).toISOString(),
+              ...(page!.lastModified ? { lastModified: page!.lastModified } : {}),
+              parserVersion: VERTEX_DOCS_PARSER_VERSION,
+              confidence: "verified",
+              fields: metadataFields(docsMetadata),
+            },
+          },
+        }
+      : staleMetadata;
+    if (!metadata) return model;
+
+    return {
+      ...metadata,
+      ...model,
+      metadataProvenance: metadata.metadataProvenance,
+    } as T;
+  });
+}
+
+/** Test seam for deterministic cache and stale-fallback assertions. */
+export function clearVertexModelMetadataCache(): void {
+  docsCache.clear();
+  docsInflight.clear();
+}
diff --git a/src/lib/providerModels/vertexPublisherModelsParser.ts b/src/lib/providerModels/vertexPublisherModelsParser.ts
new file mode 100644
index 00000000000..ea23e77472c
--- /dev/null
+++ b/src/lib/providerModels/vertexPublisherModelsParser.ts
@@ -0,0 +1,114 @@
+import {
+  getVertexModelTargetFormat,
+  normalizeVertexModelId,
+} from "@omniroute/open-sse/config/vertexModels.ts";
+
+interface VertexPublisherModel {
+  name?: string;
+  id?: string;
+  displayName?: string;
+  description?: string;
+  supportedActions?: Record<string, unknown>;
+}
+
+export interface VertexPublisherDiscoveryModel {
+  id: string;
+  name: string;
+  supportedEndpoints: ["chat"];
+  targetFormat?: "claude" | "openai";
+  owned_by: string;
+  description?: string;
+}
+
+function isCurrentGeminiChatModel(id: string): boolean {
+  const match = id.match(/^gemini-(\d+(?:\.\d+)?)-(flash-lite|flash|pro)(?:-preview)?$/i);
+  return !!match && Number(match[1]) >= 2.5;
+}
+
+function readPublisherModelId(model: VertexPublisherModel, publisher: string): string | null {
+  const rawId =
+    (typeof model.name === "string" && model.name) ||
+    (typeof model.id === "string" && model.id) ||
+    "";
+  if (!rawId) return null;
+  return rawId.includes("/") ? rawId : `publishers/${publisher}/models/${rawId}`;
+}
+
+function isPublisherChatAction(actions: Record<string, unknown> | undefined): boolean {
+  if (!actions) return true;
+  return !!(
+    actions.viewRestApi ||
+    actions.openGenerationAiStudio ||
+    actions.openGenie ||
+    actions.requestAccess
+  );
+}
+
+function toDiscoveryModel(
+  model: VertexPublisherModel,
+  id: string,
+  publisher: string,
+  targetFormat?: "claude" | "openai"
+): VertexPublisherDiscoveryModel {
+  return {
+    id,
+    name: (typeof model.displayName === "string" && model.displayName) || id,
+    supportedEndpoints: ["chat"],
+    ...(targetFormat ? { targetFormat } : {}),
+    owned_by: publisher,
+    ...(typeof model.description === "string" ? { description: model.description } : {}),
+  };
+}
+
+function toGoogleChatModel(
+  model: VertexPublisherModel,
+  routableId: string,
+  publisher: string
+): VertexPublisherDiscoveryModel | null {
+  const id = normalizeVertexModelId(routableId);
+  if (!isCurrentGeminiChatModel(id)) return null;
+  return toDiscoveryModel(model, id, publisher);
+}
+
+function toPartnerChatModel(
+  model: VertexPublisherModel,
+  routableId: string,
+  publisher: string
+): VertexPublisherDiscoveryModel | null {
+  if (!isPublisherChatAction(model.supportedActions)) return null;
+  const id = normalizeVertexModelId(routableId);
+  const targetFormat = getVertexModelTargetFormat(routableId);
+  if (!id || !targetFormat || /(?:^|[-/])ocr(?:-|$)/i.test(id)) return null;
+  return toDiscoveryModel(model, id, publisher, targetFormat);
+}
+
+function toPublisherDiscoveryModel(
+  candidate: unknown,
+  publisher: string
+): VertexPublisherDiscoveryModel[] {
+  if (!candidate || typeof candidate !== "object" || Array.isArray(candidate)) return [];
+  const model = candidate as VertexPublisherModel;
+  const routableId = readPublisherModelId(model, publisher);
+  if (!routableId) return [];
+  const parsed =
+    publisher.toLowerCase() === "google"
+      ? toGoogleChatModel(model, routableId, publisher)
+      : toPartnerChatModel(model, routableId, publisher);
+  return parsed ? [parsed] : [];
+}
+
+/** Parse one Model Garden publisher-list envelope into model ids accepted by Vertex inference. */
+export function parseVertexPublisherModels(
+  data: unknown,
+  publisher: string
+): VertexPublisherDiscoveryModel[] {
+  if (!data || typeof data !== "object" || Array.isArray(data)) return [];
+  const record = data as { models?: unknown[]; publisherModels?: unknown[] };
+  const models = Array.isArray(record.publisherModels)
+    ? record.publisherModels
+    : Array.isArray(record.models)
+      ? record.models
+      : [];
+
+  return models.flatMap((candidate) => toPublisherDiscoveryModel(candidate, publisher));
+}
diff --git a/src/lib/providerModels/vertexXaiModelsParser.ts b/src/lib/providerModels/vertexXaiModelsParser.ts
new file mode 100644
index 00000000000..e1929674633
--- /dev/null
+++ b/src/lib/providerModels/vertexXaiModelsParser.ts
@@ -0,0 +1,28 @@
+import { isVertexXaiModel } from "@omniroute/open-sse/config/vertexModels.ts";
+
+import { parseVertexPublisherModels } from "./vertexPublisherModelsParser";
+
+export interface VertexXaiDiscoveryModel {
+  id: string;
+  name: string;
+  supportedEndpoints: string[];
+  targetFormat: "openai";
+  owned_by: "xai";
+  description?: string;
+}
+
+/** Parse the v1beta1 Model Garden xAI publisher list into routable MaaS model rows. */
+export function parseVertexXaiModels(data: unknown): VertexXaiDiscoveryModel[] {
+  return parseVertexPublisherModels(data, "xai").flatMap((model) =>
+    isVertexXaiModel(model.id)
+      ? [
+          {
+            ...model,
+            supportedEndpoints: ["chat"],
+            targetFormat: "openai" as const,
+            owned_by: "xai" as const,
+          },
+        ]
+      : []
+  );
+}
diff --git a/src/shared/constants/providers/apikey/enterprise-cloud.ts b/src/shared/constants/providers/apikey/enterprise-cloud.ts
index 49831b3c240..9d65a55636b 100644
--- a/src/shared/constants/providers/apikey/enterprise-cloud.ts
+++ b/src/shared/constants/providers/apikey/enterprise-cloud.ts
@@ -119,7 +119,8 @@ export const APIKEY_PROVIDERS_ENTERPRISE = {
     textIcon: "VA",
     website: "https://cloud.google.com/vertex-ai",
     hasFree: true,
-    authHint: "Provide Service Account JSON or OAuth access_token",
+    authHint:
+      "Provide Service Account JSON, an OAuth access token, a Vertex Express API key, or a service-account-bound authorization key. Express mode supports Gemini only; partner models require project-scoped credentials.",
   },
   "vertex-partner": {
     id: "vertex-partner",
@@ -130,7 +131,8 @@ export const APIKEY_PROVIDERS_ENTERPRISE = {
     color: "#34A853",
     textIcon: "VP",
     website: "https://cloud.google.com/vertex-ai",
-    authHint: "Provide the same Service Account JSON used for Vertex AI partner models.",
+    authHint:
+      "Provide Service Account JSON or OAuth credentials. A service-account-bound authorization key also supports discovery, but partner inference additionally requires its Google Cloud project ID. Standard Express keys support Gemini only.",
   },
   "cloudflare-ai": {
     id: "cloudflare-ai",
diff --git a/tests/unit/credential-hint-copy-3180-3091.test.ts b/tests/unit/credential-hint-copy-3180-3091.test.ts
index e3ae205fd9e..1b502c2de22 100644
--- a/tests/unit/credential-hint-copy-3180-3091.test.ts
+++ b/tests/unit/credential-hint-copy-3180-3091.test.ts
@@ -25,8 +25,12 @@ test("#3180 grok-web credential hint names both sso and sso-rw", () => {
 
 test("#3091 vertex Service Account placeholder is real instructional text, not the stub", () => {
   const en = JSON.parse(readFileSync(path.join(MESSAGES_DIR, "en.json"), "utf8"));
+  const de = JSON.parse(readFileSync(path.join(MESSAGES_DIR, "de.json"), "utf8"));
   const placeholder = en.providers?.vertexServiceAccountPlaceholder;
   assert.equal(typeof placeholder, "string");
   assert.notEqual(placeholder, "Vertex Service Account Placeholder");
   assert.match(placeholder, /service_account/);
+  assert.match(en.providers?.vertexCredentialHint || "", /live Model Garden discovery/i);
+  assert.match(de.providers?.vertexCredentialLabel || "", /Service-Account-JSON/);
+  assert.match(de.providers?.vertexCredentialHint || "", /Live-Erkennung/);
 });
diff --git a/tests/unit/executor-vertex-extended.test.ts b/tests/unit/executor-vertex-extended.test.ts
index d0093911b29..4efc1cac998 100644
--- a/tests/unit/executor-vertex-extended.test.ts
+++ b/tests/unit/executor-vertex-extended.test.ts
@@ -68,7 +68,87 @@ test("VertexExecutor.buildUrl routes a non-JSON Express API key to the project-l
     expressUrl,
     "https://aiplatform.googleapis.com/v1/publishers/google/models/gemini-2.5-flash:generateContent?key=express-key-abc"
   );
-  assert.ok(!expressUrl.includes("/projects/"), "Express key URL must not route through a project path");
+  assert.ok(
+    !expressUrl.includes("/projects/"),
+    "Express key URL must not route through a project path"
+  );
+});
+
+test("VertexExecutor.buildUrl rejects partner models for project-less Express credentials", () => {
+  const executor = new VertexExecutor();
+  const ids = ["grok-4.6", "xai/grok-4.6", "xai/models/grok-4.6", "publishers/xai/models/grok-4.6"];
+
+  for (const modelId of ids) {
+    assert.throws(
+      () => executor.buildUrl(modelId, false, 0, { apiKey: "k-express" }),
+      /partner models require project-scoped credentials/i,
+      modelId
+    );
+  }
+});
+
+test("VertexExecutor.execute canonicalizes an xAI resource id in URL and request body", async () => {
+  const executor = new VertexExecutor();
+  const originalFetch = globalThis.fetch;
+  const calls: Array<{ url: string; body: string }> = [];
+
+  globalThis.fetch = async (url: RequestInfo | URL, init?: RequestInit) => {
+    calls.push({ url: String(url), body: String(init?.body || "") });
+    return Response.json({ choices: [] });
+  };
+
+  try {
+    await executor.execute({
+      model: "publishers/xai/models/grok-4.6",
+      body: {
+        model: "publishers/xai/models/grok-4.6",
+        messages: [{ role: "user", content: "hi" }],
+      },
+      stream: false,
+      credentials: {
+        apiKey: "k-authorization",
+        projectId: "proj-xai",
+      },
+    });
+
+    assert.equal(calls.length, 1);
+    assert.equal(
+      calls[0].url,
+      "https://aiplatform.googleapis.com/v1/projects/proj-xai/locations/global/endpoints/openapi/chat/completions?key=k-authorization"
+    );
+    assert.equal(JSON.parse(calls[0].body).model, "xai/grok-4.6");
+  } finally {
+    globalThis.fetch = originalFetch;
+  }
+});
+
+test("VertexExecutor.buildUrl routes Mistral Model Garden ids to native rawPredict", () => {
+  const executor = new VertexExecutor();
+  const credentials = {
+    apiKey: createServiceAccountJson({ projectId: "proj-mistral" }),
+    providerSpecificData: { region: "europe-west4" },
+  };
+
+  assert.equal(
+    executor.buildUrl("publishers/mistralai/models/mistral-medium-3", false, 0, credentials),
+    "https://aiplatform.googleapis.com/v1/projects/proj-mistral/locations/europe-west4/publishers/mistralai/models/mistral-medium-3:rawPredict"
+  );
+  assert.equal(
+    executor.buildUrl("mistralai/mistral-medium-3", true, 0, credentials),
+    "https://aiplatform.googleapis.com/v1/projects/proj-mistral/locations/europe-west4/publishers/mistralai/models/mistral-medium-3:streamRawPredict"
+  );
+});
+
+test("VertexExecutor.buildUrl generically routes future publisher resources to OpenAI MaaS", () => {
+  const executor = new VertexExecutor();
+  const url = executor.buildUrl("publishers/future-vendor/models/future-chat-maas", false, 0, {
+    apiKey: createServiceAccountJson({ projectId: "proj-future" }),
+  });
+
+  assert.equal(
+    url,
+    "https://aiplatform.googleapis.com/v1/projects/proj-future/locations/global/endpoints/openapi/chat/completions"
+  );
 });
 
 test("VertexExecutor.buildUrl routes partner and org-prefixed models to the global partner endpoint", () => {
@@ -79,6 +159,9 @@ test("VertexExecutor.buildUrl routes partner and org-prefixed models to the glob
   const metaLlama = executor.buildUrl("meta/llama-3.1-405b-instruct-maas", true, 0, {
     apiKey: createServiceAccountJson({ projectId: "proj-llama" }),
   });
+  const grok = executor.buildUrl("publishers/xai/models/grok-4.6", true, 0, {
+    apiKey: createServiceAccountJson({ projectId: "proj-xai" }),
+  });
 
   assert.equal(
     deepseek,
@@ -88,6 +171,36 @@ test("VertexExecutor.buildUrl routes partner and org-prefixed models to the glob
     metaLlama,
     "https://aiplatform.googleapis.com/v1/projects/proj-llama/locations/global/endpoints/openapi/chat/completions"
   );
+  assert.equal(
+    grok,
+    "https://aiplatform.googleapis.com/v1/projects/proj-xai/locations/global/endpoints/openapi/chat/completions"
+  );
+});
+
+test("VertexExecutor.execute namespaces legacy bare open-MaaS model ids", async () => {
+  const executor = new VertexExecutor();
+  const originalFetch = globalThis.fetch;
+  let sentModel: string | undefined;
+
+  globalThis.fetch = async (_url, init) => {
+    sentModel = JSON.parse(String(init?.body)).model;
+    return Response.json({ choices: [] });
+  };
+
+  try {
+    await executor.execute({
+      model: "DeepSeek-V4-Pro",
+      body: { model: "DeepSeek-V4-Pro", messages: [{ role: "user", content: "hi" }] },
+      stream: false,
+      credentials: {
+        apiKey: createServiceAccountJson({ projectId: "proj-deepseek" }),
+        accessToken: "ya29.deepseek",
+      },
+    });
+    assert.equal(sentModel, "deepseek-ai/DeepSeek-V4-Pro");
+  } finally {
+    globalThis.fetch = originalFetch;
+  }
 });
 
 test("VertexExecutor.buildUrl routes current-generation Claude models to the native Anthropic rawPredict endpoint (#1985, #8994)", () => {
diff --git a/tests/unit/t28-model-catalog-updates.test.ts b/tests/unit/t28-model-catalog-updates.test.ts
index 5a3c8c129bb..84a22e85867 100644
--- a/tests/unit/t28-model-catalog-updates.test.ts
+++ b/tests/unit/t28-model-catalog-updates.test.ts
@@ -113,11 +113,35 @@ test("T28: lmarena registry seeds Direct-chat Text/search; image models in IMAGE
 
 test("T28: vertex catalog includes partner models when vertex executor is available", () => {
   const vertexIds = REGISTRY.vertex.models.map((m) => m.id);
+  const vertexBudgetIds = FREE_MODEL_BUDGETS.filter((m) => m.provider === "vertex").map(
+    (m) => m.modelId
+  );
 
-  assert.ok(vertexIds.includes("DeepSeek-V4-Flash"));
-  assert.ok(vertexIds.includes("DeepSeek-V4-Pro"));
-  assert.ok(vertexIds.includes("Qwen3.6-35B-A3B"));
-  assert.ok(vertexIds.includes("GLM-5.1-FP8"));
+  assert.ok(vertexIds.includes("gemini-3.7-flash"));
+  assert.ok(vertexIds.includes("gemini-3.6-flash"));
+  assert.ok(vertexIds.includes("gemini-3.5-flash"));
+  assert.ok(vertexIds.includes("gemini-3.5-flash-lite"));
+  assert.ok(vertexIds.includes("gemini-2.5-pro"));
+  assert.ok(vertexIds.includes("gemini-2.5-flash"));
+  assert.ok(vertexIds.includes("gemini-2.5-flash-lite"));
+  assert.ok(!vertexIds.includes("gemma-4-31b-it"));
+  assert.ok(vertexIds.includes("deepseek-ai/deepseek-v3.2-maas"));
+  assert.ok(vertexIds.includes("deepseek-ai/deepseek-v3.1-maas"));
+  assert.ok(vertexIds.includes("qwen/qwen3-next-80b-a3b-instruct-maas"));
+  assert.ok(vertexIds.includes("zai-org/glm-5-maas"));
+  assert.ok(!vertexIds.includes("DeepSeek-V4-Flash"));
+  assert.ok(!vertexIds.includes("DeepSeek-V4-Pro"));
+  assert.ok(!vertexIds.includes("Qwen3.6-35B-A3B"));
+  assert.ok(!vertexIds.includes("GLM-5.1-FP8"));
+  assert.ok(vertexBudgetIds.includes("deepseek-ai/deepseek-v3.2-maas"));
+  assert.ok(vertexBudgetIds.includes("gemini-3.7-flash"));
+  assert.ok(vertexBudgetIds.includes("gemini-3.6-flash"));
+  assert.ok(!vertexBudgetIds.includes("gemma-4-31b-it"));
+  assert.ok(vertexBudgetIds.includes("qwen/qwen3-next-80b-a3b-instruct-maas"));
+  assert.ok(vertexBudgetIds.includes("zai-org/glm-5-maas"));
+  assert.ok(!vertexBudgetIds.includes("DeepSeek-V4-Flash"));
+  assert.ok(!vertexBudgetIds.includes("Qwen3.6-35B-A3B"));
+  assert.ok(!vertexBudgetIds.includes("GLM-5.1-FP8"));
 });
 
 test("T28: volcengine (Ark) catalog includes DeepSeek V4 models", () => {
@@ -151,7 +175,7 @@ test("T28: new catalog models resolve through getModelInfoCore", async () => {
   assert.equal(flashPreview.provider, "gemini");
   assert.equal(flashPreview.model, "gemini-3-flash-preview");
 
-  const vertexPartner = await getModelInfoCore("vertex/Qwen3.6-35B-A3B", {});
+  const vertexPartner = await getModelInfoCore("vertex/qwen/qwen3-next-80b-a3b-instruct-maas", {});
   assert.equal(vertexPartner.provider, "vertex");
-  assert.equal(vertexPartner.model, "Qwen3.6-35B-A3B");
+  assert.equal(vertexPartner.model, "qwen/qwen3-next-80b-a3b-instruct-maas");
 });
diff --git a/tests/unit/t29-vertex-sa-json-executor.test.ts b/tests/unit/t29-vertex-sa-json-executor.test.ts
index f367718084e..ffa0dc25413 100644
--- a/tests/unit/t29-vertex-sa-json-executor.test.ts
+++ b/tests/unit/t29-vertex-sa-json-executor.test.ts
@@ -1,7 +1,7 @@
 import test from "node:test";
 import assert from "node:assert/strict";
 
-const { VertexExecutor } = await import("../../open-sse/executors/vertex.ts");
+const { VertexExecutor, parseSAFromApiKey } = await import("../../open-sse/executors/vertex.ts");
 
 const MIN_SA_JSON = JSON.stringify({
   project_id: "vertex-project-123",
@@ -22,7 +22,7 @@ test("T29: Vertex executor builds regional Gemini URL from Service Account proje
 
 test("T29: Vertex executor routes partner models to global openapi endpoint", () => {
   const executor = new VertexExecutor();
-  const url = executor.buildUrl("DeepSeek-V4-Pro", false, 0, {
+  const url = executor.buildUrl("deepseek-ai/deepseek-v3.2-maas", false, 0, {
     apiKey: MIN_SA_JSON,
     providerSpecificData: { region: "us-central1" },
   });
@@ -72,6 +72,16 @@ test("T29: Vertex executor rejects incomplete Service Account JSON clearly", asy
   );
 });
 
+test("T29: Vertex distinguishes OAuth client JSON from a Service Account key", () => {
+  assert.throws(
+    () =>
+      parseSAFromApiKey(
+        JSON.stringify({ web: { client_id: "example.apps.googleusercontent.com" } })
+      ),
+    /OAuth client JSON is not a Service Account key/i
+  );
+});
+
 test("T29: Vertex executor routes a non-JSON Express API key to the project-less publisher endpoint", () => {
   const executor = new VertexExecutor();
   const stream = executor.buildUrl("gemini-2.5-flash", true, 0, { apiKey: "express-key-123" });
diff --git a/tests/unit/validation-specialty-inline-split.test.ts b/tests/unit/validation-specialty-inline-split.test.ts
index 39c734c961e..6576a9463de 100644
--- a/tests/unit/validation-specialty-inline-split.test.ts
+++ b/tests/unit/validation-specialty-inline-split.test.ts
@@ -67,3 +67,11 @@ test("validateVertexProvider: malformed Service Account JSON is rejected", async
   assert.equal(result.valid, false);
   assert.match(result.error || "", /Invalid Service Account JSON/i);
 });
+
+test("validateVertexProvider: OAuth client JSON explains the required credential type", async () => {
+  const result = await M.validateVertexProvider({
+    apiKey: JSON.stringify({ web: { client_id: "example.apps.googleusercontent.com" } }),
+  });
+  assert.equal(result.valid, false);
+  assert.match(result.error || "", /OAuth client JSON is not a Service Account key/i);
+});
diff --git a/tests/unit/vertex-model-metadata.test.ts b/tests/unit/vertex-model-metadata.test.ts
new file mode 100644
index 00000000000..5752d945561
--- /dev/null
+++ b/tests/unit/vertex-model-metadata.test.ts
@@ -0,0 +1,165 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+
+import { normalizeDiscoveredModels } from "../../src/lib/providerModels/modelDiscovery.ts";
+import {
+  clearVertexModelMetadataCache,
+  enrichVertexModelsWithMetadata,
+  parseVertexModelDocsHtml,
+  resolveVertexModelDocsUrls,
+} from "../../src/lib/providerModels/vertexModelMetadata.ts";
+
+function modelDocsHtml(rows: string): string {
+  return `<html><body><table><tbody>${rows}</tbody></table></body></html>`;
+}
+
+test.beforeEach(() => clearVertexModelMetadataCache());
+
+test("Vertex docs resolver maps native and MaaS model IDs only to the official docs host", () => {
+  assert.deepEqual(resolveVertexModelDocsUrls("gemini-3.7-flash"), [
+    "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-7-flash?hl=en",
+  ]);
+  assert.deepEqual(resolveVertexModelDocsUrls("claude-opus-4-1"), [
+    "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/claude/opus-4-1?hl=en",
+  ]);
+  assert.deepEqual(resolveVertexModelDocsUrls("xai/grok-4.20-reasoning"), [
+    "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/grok/grok-4-20?hl=en",
+  ]);
+  assert.deepEqual(resolveVertexModelDocsUrls("deepseek-ai/deepseek-v3.2-maas"), [
+    "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/maas/deepseek-ai/deepseek-v3-2-maas?hl=en",
+    "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/maas/deepseek-ai/deepseek-v3-2?hl=en",
+  ]);
+  assert.ok(
+    resolveVertexModelDocsUrls("xai/grok-4.6").every((url) =>
+      url.startsWith("https://docs.cloud.google.com/")
+    )
+  );
+});
+
+test("Vertex docs parser extracts distinct context, input, and output limits", () => {
+  const gemini = parseVertexModelDocsHtml(
+    modelDocsHtml(`
+      <tr><th>Model ID</th><td><code>gemini-3.7-flash</code></td></tr>
+      <tr><th>Token limits</th><td>Context window</td><td>1,048,576</td></tr>
+      <tr><th>Maximum output tokens</th><td>65,536</td></tr>
+    `),
+    "gemini-3.7-flash"
+  );
+  assert.deepEqual(gemini, { contextWindow: 1048576, outputTokenLimit: 65536 });
+
+  const claude = parseVertexModelDocsHtml(
+    modelDocsHtml(`
+      <tr><th>Model ID</th><td>claude-opus-4-1</td></tr>
+      <tr><th>Token limits</th><td>
+        Maximum input tokens: 200,000 Maximum output tokens: 32,000
+      </td></tr>
+      <tr><th>Quota limits</th><td>Context length: 200,000</td></tr>
+    `),
+    "claude-opus-4-1"
+  );
+  assert.deepEqual(claude, {
+    contextWindow: 200000,
+    inputTokenLimit: 200000,
+    outputTokenLimit: 32000,
+  });
+});
+
+test("Vertex docs parser rejects a page without the exact live model ID", () => {
+  const html = modelDocsHtml(`
+    <tr><th>Model ID</th><td>gemini-3.7-flash-lite</td></tr>
+    <tr><th>Token limits</th><td>Context window</td><td>1,048,576</td></tr>
+  `);
+  assert.equal(parseVertexModelDocsHtml(html, "gemini-3.7-flash"), null);
+});
+
+test("Vertex enrichment deduplicates shared docs fetches and preserves structured API limits", async () => {
+  const html = modelDocsHtml(`
+    <tr><th>Model ID</th><td>grok-4.20-reasoning</td></tr>
+    <tr><th>Model ID</th><td>grok-4.20-non-reasoning</td></tr>
+    <tr><th>Quota limits</th><td>Context length: 2,000,000</td></tr>
+    <tr><th>Maximum output tokens</th><td>80,000</td></tr>
+  `);
+  let fetchCalls = 0;
+  const models = await enrichVertexModelsWithMetadata({
+    models: [
+      { id: "xai/grok-4.20-reasoning", name: "Reasoning", outputTokenLimit: 12345 },
+      { id: "xai/grok-4.20-non-reasoning", name: "Non-Reasoning" },
+    ],
+    now: new Date("2026-09-04T12:00:00.000Z"),
+    fetchImpl: async () => {
+      fetchCalls += 1;
+      return new Response(html, {
+        status: 200,
+        headers: { "Last-Modified": "Thu, 03 Sep 2026 10:00:00 GMT" },
+      });
+    },
+  });
+
+  assert.equal(fetchCalls, 1);
+  assert.equal(models[0]?.contextWindow, 2000000);
+  assert.equal(models[0]?.outputTokenLimit, 12345);
+  assert.equal(models[1]?.outputTokenLimit, 80000);
+  assert.deepEqual(models[1]?.metadataProvenance?.vertexDocs.fields, [
+    "contextWindow",
+    "outputTokenLimit",
+  ]);
+  assert.equal(
+    models[1]?.metadataProvenance?.vertexDocs.sourceUrl,
+    "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/grok/grok-4-20?hl=en"
+  );
+});
+
+test("Vertex enrichment reuses only recently verified persisted metadata on docs failure", async () => {
+  const staleModel = {
+    id: "gemini-3.7-flash",
+    contextWindow: 1048576,
+    outputTokenLimit: 65536,
+    metadataProvenance: {
+      vertexDocs: {
+        source: "google-cloud-docs" as const,
+        sourceUrl:
+          "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-7-flash?hl=en",
+        fetchedAt: "2026-09-01T12:00:00.000Z",
+        parserVersion: "vertex-docs-v1" as const,
+        confidence: "verified" as const,
+        fields: ["contextWindow", "outputTokenLimit"] as const,
+      },
+    },
+  };
+  const [withinTtl] = await enrichVertexModelsWithMetadata({
+    models: [{ id: staleModel.id }],
+    staleModels: [staleModel],
+    now: new Date("2026-09-04T12:00:00.000Z"),
+    fetchImpl: async () => new Response("unavailable", { status: 503 }),
+  });
+  assert.equal(withinTtl?.contextWindow, 1048576);
+  assert.equal(withinTtl?.outputTokenLimit, 65536);
+
+  clearVertexModelMetadataCache();
+  const [expired] = await enrichVertexModelsWithMetadata({
+    models: [{ id: staleModel.id }],
+    staleModels: [staleModel],
+    now: new Date("2026-09-10T12:00:00.001Z"),
+    fetchImpl: async () => new Response("unavailable", { status: 503 }),
+  });
+  assert.equal(expired?.contextWindow, undefined);
+  assert.equal(expired?.outputTokenLimit, undefined);
+});
+
+test("Vertex normalization persists context separately while legacy providers retain compatibility", () => {
+  const metadata = {
+    id: "gemini-3.7-flash",
+    contextWindow: 1048576,
+    outputTokenLimit: 65536,
+  };
+  assert.deepEqual(normalizeDiscoveredModels([metadata], "vertex"), [
+    {
+      id: "gemini-3.7-flash",
+      name: "gemini-3.7-flash",
+      source: "imported",
+      contextWindow: 1048576,
+      outputTokenLimit: 65536,
+    },
+  ]);
+  assert.equal(normalizeDiscoveredModels([metadata], "openrouter")[0]?.inputTokenLimit, 1048576);
+});
diff --git a/tests/unit/vertex-models-route.test.ts b/tests/unit/vertex-models-route.test.ts
new file mode 100644
index 00000000000..e89d041bbb5
--- /dev/null
+++ b/tests/unit/vertex-models-route.test.ts
@@ -0,0 +1,278 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-vertex-model-routes-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+
+const core = await import("../../src/lib/db/core.ts");
+const providersDb = await import("../../src/lib/db/providers.ts");
+const providerModelsRoute = await import("../../src/app/api/providers/[id]/models/route.ts");
+
+const originalFetch = globalThis.fetch;
+
+async function resetStorage(): Promise<void> {
+  globalThis.fetch = originalFetch;
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 });
+  fs.mkdirSync(TEST_DATA_DIR, { recursive: true });
+}
+
+async function seedVertexConnection(options: {
+  apiKey?: string;
+  accessToken?: string;
+  projectId?: string;
+}): Promise<{ id: string }> {
+  return providersDb.createProviderConnection({
+    provider: "vertex",
+    authType: "apikey",
+    name: `vertex-${Math.random().toString(16).slice(2, 8)}`,
+    apiKey: options.apiKey,
+    accessToken: options.accessToken,
+    projectId: options.projectId,
+    isActive: true,
+    testStatus: "active",
+    providerSpecificData: { autoFetchModels: true },
+  });
+}
+
+async function callRoute(connectionId: string): Promise<Response> {
+  return providerModelsRoute.GET(
+    new Request(`http://localhost/api/providers/${connectionId}/models?refresh=true&chatOnly=true`),
+    { params: { id: connectionId } }
+  );
+}
+
+test.beforeEach(async () => {
+  await resetStorage();
+});
+
+test.after(async () => {
+  globalThis.fetch = originalFetch;
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 });
+});
+
+test("Vertex Express probes only Gemini discovery and then uses the intentional Gemini catalog", async () => {
+  const connection = await seedVertexConnection({ apiKey: "vertex-express-key" });
+  let fetchCalls = 0;
+  const headers: HeadersInit[] = [];
+  globalThis.fetch = async (_url, init) => {
+    fetchCalls += 1;
+    headers.push(init?.headers || {});
+    return Response.json({ error: { status: "UNAUTHENTICATED" } }, { status: 401 });
+  };
+
+  const response = await callRoute(connection.id);
+  const body = await response.json();
+
+  assert.equal(response.status, 200);
+  assert.equal(body.source, "local_catalog");
+  assert.equal(body.intentional, true);
+  assert.match(body.warning, /live catalog.*API key.*curated Express catalog/i);
+  assert.equal(fetchCalls, 1);
+  assert.ok(
+    headers.every((header) => new Headers(header).get("x-goog-api-key") === "vertex-express-key")
+  );
+  assert.ok(body.models.some((model: { id?: string }) => model.id.startsWith("gemini-")));
+  assert.ok(!body.models.some((model: { id?: string }) => model.id.includes("grok")));
+});
+
+test("Vertex API-key endpoint rejection remains intentional and names the HTTP failure", async () => {
+  for (const status of [400, 403]) {
+    const connection = await seedVertexConnection({ apiKey: "vertex-express-key" });
+    globalThis.fetch = async () => Response.json({ error: { message: "Rejected" } }, { status });
+    const response = await callRoute(connection.id);
+    const body = await response.json();
+    assert.equal(response.status, 200);
+    assert.equal(body.intentional, true);
+    assert.match(body.warning, new RegExp("HTTP " + status));
+  }
+});
+
+test("Vertex transient HTTP and network failures do not become intentional catalogs", async () => {
+  for (const status of [429, 500, 503, 0]) {
+    const connection = await seedVertexConnection({ apiKey: "vertex-express-key" });
+    globalThis.fetch = async () => {
+      if (!status) throw new Error("Fixture network failure");
+      return Response.json({ error: { message: "Unavailable" } }, { status });
+    };
+    const response = await callRoute(connection.id);
+    const body = await response.json();
+    assert.equal(response.status, 200);
+    assert.equal(body.intentional, undefined);
+    assert.match(body.warning, /temporarily unavailable/);
+  }
+});
+
+test("Vertex authorization API key detects and persists its project for curated partner models", async () => {
+  const connection = await seedVertexConnection({ apiKey: "vertex-authorization-key" });
+  const calledUrls: string[] = [];
+  globalThis.fetch = async (url, init) => {
+    const calledUrl = String(url);
+    calledUrls.push(calledUrl);
+    assert.equal(new Headers(init?.headers).get("x-goog-api-key"), "vertex-authorization-key");
+    return Response.json(
+      {
+        error: {
+          status: "PERMISSION_DENIED",
+          details: [
+            {
+              "@type": "type.googleapis.com/google.rpc.ErrorInfo",
+              reason: "API_KEY_SERVICE_BLOCKED",
+              metadata: { consumer: "projects/316081256616" },
+            },
+          ],
+        },
+      },
+      { status: 403 }
+    );
+  };
+
+  const response = await callRoute(connection.id);
+  const body = await response.json();
+  const byId = new Map(
+    body.models.map((model: { id: string; targetFormat?: string }) => [model.id, model])
+  );
+  assert.equal(response.status, 200);
+  assert.equal(body.source, "local_catalog");
+  assert.equal(body.intentional, true);
+  assert.equal(body.projectIdAutoDetected, true);
+  assert.equal(body.catalogMode, "curated_project");
+  assert.equal(body.warning, undefined);
+  assert.equal(byId.get("xai/grok-4.6")?.targetFormat, "openai");
+  assert.equal(byId.get("zai-org/glm-5-maas")?.targetFormat, "openai");
+  assert.equal(byId.get("qwen/qwen3-next-80b-a3b-instruct-maas")?.targetFormat, "openai");
+  assert.ok(!byId.has("GLM-5.1-FP8"));
+  assert.ok(!byId.has("Qwen3.6-35B-A3B"));
+  assert.equal(calledUrls.length, 1);
+  assert.ok(calledUrls[0].includes("generativelanguage.googleapis.com"));
+  assert.ok(!calledUrls[0].includes("vertex-authorization-key"));
+
+  const saved = await providersDb.getProviderConnectionById(connection.id);
+  assert.equal(saved?.projectId, "316081256616");
+});
+
+test("Vertex Service Account discovery merges Gemini and all partner transport catalogs", async () => {
+  const connection = await seedVertexConnection({
+    accessToken: "ya29.vertex-model-discovery",
+  });
+  const calledUrls: string[] = [];
+  globalThis.fetch = async (url) => {
+    const calledUrl = String(url);
+    calledUrls.push(calledUrl);
+    if (calledUrl.includes("generativelanguage.googleapis.com")) {
+      return Response.json({
+        models: [
+          {
+            name: "models/gemini-3.1-pro-preview",
+            displayName: "Gemini 3.1 Pro Preview",
+            supportedGenerationMethods: ["generateContent"],
+          },
+          {
+            name: "models/gemini-3.7-flash",
+            displayName: "Gemini 3.7 Flash",
+            supportedGenerationMethods: ["generateContent"],
+            inputTokenLimit: 900000,
+            outputTokenLimit: 60000,
+          },
+        ],
+      });
+    }
+    if (calledUrl.includes("/models/gemini/3-7-flash?hl=en")) {
+      return new Response(
+        `<table>
+          <tr><th>Model ID</th><td>gemini-3.7-flash</td></tr>
+          <tr><th>Token limits</th><td>Context window</td><td>1,048,576</td></tr>
+          <tr><th>Maximum output tokens</th><td>65,536</td></tr>
+        </table>`,
+        { headers: { "Last-Modified": "Thu, 03 Sep 2026 10:00:00 GMT" } }
+      );
+    }
+    if (calledUrl.includes("/publishers/google/models")) {
+      return Response.json({
+        publisherModels: [
+          { name: "publishers/google/models/gemini-3.7-flash" },
+          { name: "publishers/google/models/gemini-3.1-flash-image" },
+        ],
+      });
+    }
+    if (calledUrl.includes("/publishers/anthropic/models")) {
+      return Response.json({
+        publisherModels: [{ name: "publishers/anthropic/models/claude-sonnet-4-6" }],
+      });
+    }
+    if (calledUrl.includes("/publishers/xai/models")) {
+      return Response.json({
+        publisherModels: [{ name: "publishers/xai/models/grok-4.6" }],
+      });
+    }
+    if (calledUrl.includes("/publishers/mistralai/models")) {
+      return Response.json({
+        publisherModels: [{ name: "publishers/mistralai/models/mistral-medium-3" }],
+      });
+    }
+    return new Response("unexpected discovery URL", { status: 500 });
+  };
+
+  const response = await callRoute(connection.id);
+  const body = await response.json();
+  const byId = new Map(
+    body.models.map((model: { id: string; targetFormat?: string }) => [model.id, model])
+  );
+  const gemini37 = byId.get("gemini-3.7-flash") as
+    | {
+        contextWindow?: number;
+        inputTokenLimit?: number;
+        outputTokenLimit?: number;
+        metadataProvenance?: { vertexDocs?: { source?: string } };
+      }
+    | undefined;
+
+  assert.equal(response.status, 200);
+  assert.equal(body.source, "api");
+  assert.ok(byId.has("gemini-3.1-pro-preview"));
+  assert.ok(byId.has("gemini-3.7-flash"));
+  assert.equal(gemini37?.contextWindow, 1048576);
+  assert.equal(gemini37?.inputTokenLimit, 900000);
+  assert.equal(gemini37?.outputTokenLimit, 60000);
+  assert.equal(gemini37?.metadataProvenance?.vertexDocs?.source, "google-cloud-docs");
+  assert.ok(!byId.has("gemini-3.1-flash-image"));
+  assert.equal(byId.get("claude-sonnet-4-6")?.targetFormat, "claude");
+  assert.equal(byId.get("xai/grok-4.6")?.targetFormat, "openai");
+  assert.equal(byId.get("mistral-medium-3")?.targetFormat, "openai");
+  assert.ok(calledUrls.some((url) => url.includes("/v1beta1/publishers/xai/models")));
+  assert.ok(calledUrls.some((url) => url.includes("/v1beta1/publishers/google/models")));
+  assert.ok(
+    calledUrls
+      .filter((url) => url.includes("/v1beta1/publishers/"))
+      .every((url) => url.includes("pageSize=300"))
+  );
+});
+
+test("Vertex Service Account keeps usable xAI results when Gemini listing is blocked", async () => {
+  const connection = await seedVertexConnection({ accessToken: "ya29.vertex-xai-only" });
+  globalThis.fetch = async (url) => {
+    const calledUrl = String(url);
+    if (calledUrl.includes("generativelanguage.googleapis.com")) {
+      return Response.json({ error: { status: "PERMISSION_DENIED" } }, { status: 403 });
+    }
+    if (calledUrl.includes("/publishers/xai/models")) {
+      return Response.json({
+        publisherModels: [{ name: "publishers/xai/models/grok-4.6" }],
+      });
+    }
+    return Response.json({ publisherModels: [] });
+  };
+
+  const response = await callRoute(connection.id);
+  const body = await response.json();
+
+  assert.equal(response.status, 200);
+  assert.equal(body.source, "api");
+  assert.ok(body.models.some((model: { id?: string }) => model.id === "xai/grok-4.6"));
+  assert.equal(body.catalogMode, "live_vertex_catalog");
+  assert.equal(body.warning, undefined);
+});
diff --git a/tests/unit/vertex-publisher-models-parser.test.ts b/tests/unit/vertex-publisher-models-parser.test.ts
new file mode 100644
index 00000000000..04c22e3c7e4
--- /dev/null
+++ b/tests/unit/vertex-publisher-models-parser.test.ts
@@ -0,0 +1,157 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import { discoverVertexModelsWithApiKey } from "../../src/lib/providerModels/vertexModelDiscovery.ts";
+import { parseVertexPublisherModels } from "../../src/lib/providerModels/vertexPublisherModelsParser.ts";
+
+test("generic Vertex publisher parser produces routable IDs for every transport family", () => {
+  const cases = [
+    ["anthropic", "claude-sonnet-5", "claude", "claude-sonnet-5"],
+    ["mistralai", "mistral-medium-3", "openai", "mistral-medium-3"],
+    ["xai", "grok-4.6", "openai", "xai/grok-4.6"],
+    [
+      "meta",
+      "llama-4-maverick-17b-128e-instruct-maas",
+      "openai",
+      "meta/llama-4-maverick-17b-128e-instruct-maas",
+    ],
+    ["deepseek-ai", "deepseek-v3.2-maas", "openai", "deepseek-ai/deepseek-v3.2-maas"],
+    ["future-vendor", "future-chat-maas", "openai", "future-vendor/future-chat-maas"],
+  ] as const;
+
+  for (const [publisher, rawId, targetFormat, expectedId] of cases) {
+    const models = parseVertexPublisherModels(
+      {
+        publisherModels: [
+          {
+            name: `publishers/${publisher}/models/${rawId}`,
+            displayName: `Display ${rawId}`,
+          },
+        ],
+      },
+      publisher
+    );
+
+    assert.deepEqual(models, [
+      {
+        id: expectedId,
+        name: `Display ${rawId}`,
+        supportedEndpoints: ["chat"],
+        targetFormat,
+        owned_by: publisher,
+      },
+    ]);
+  }
+});
+
+test("generic Vertex publisher parser accepts models and publisherModels envelopes", () => {
+  assert.equal(
+    parseVertexPublisherModels(
+      { models: [{ id: "publishers/qwen/models/qwen3-next-80b-a3b-instruct-maas" }] },
+      "qwen"
+    )[0]?.id,
+    "qwen/qwen3-next-80b-a3b-instruct-maas"
+  );
+  assert.deepEqual(parseVertexPublisherModels(null, "xai"), []);
+  assert.deepEqual(parseVertexPublisherModels({ publisherModels: [null, {}] }, "xai"), []);
+  assert.equal(
+    parseVertexPublisherModels({ publisherModels: [{ id: "grok-4.6" }] }, "xai")[0]?.id,
+    "xai/grok-4.6"
+  );
+  assert.deepEqual(
+    parseVertexPublisherModels(
+      {
+        publisherModels: [
+          {
+            id: "self-deploy-only",
+            supportedActions: { deploy: {} },
+          },
+        ],
+      },
+      "future-vendor"
+    ),
+    []
+  );
+  assert.equal(
+    parseVertexPublisherModels(
+      {
+        publisherModels: [
+          {
+            name: "publishers/mistralai/models/mistral-medium-3",
+            supportedActions: { requestAccess: {} },
+          },
+        ],
+      },
+      "mistralai"
+    )[0]?.id,
+    "mistral-medium-3"
+  );
+  assert.deepEqual(
+    parseVertexPublisherModels(
+      {
+        publisherModels: [
+          {
+            name: "publishers/mistralai/models/mistral-ocr-2505",
+            supportedActions: { requestAccess: {} },
+          },
+        ],
+      },
+      "mistralai"
+    ),
+    []
+  );
+});
+
+test("Google publisher discovery keeps current Gemini chat models and filters media or retired IDs", () => {
+  const models = parseVertexPublisherModels(
+    {
+      publisherModels: [
+        { name: "publishers/google/models/gemini-3.7-flash" },
+        { name: "publishers/google/models/gemini-3.8-flash" },
+        { name: "publishers/google/models/gemini-3.1-pro-preview" },
+        { name: "publishers/google/models/gemini-3.1-flash-image" },
+        { name: "publishers/google/models/gemini-2.5-pro-tts" },
+        { name: "publishers/google/models/gemini-1.5-pro-002" },
+      ],
+    },
+    "google"
+  );
+
+  assert.deepEqual(
+    models.map((model) => model.id),
+    ["gemini-3.7-flash", "gemini-3.8-flash", "gemini-3.1-pro-preview"]
+  );
+  assert.ok(models.every((model) => model.targetFormat === undefined));
+});
+
+test("Vertex API-key discovery extracts its consumer project without probing Model Garden", async () => {
+  const urls: string[] = [];
+  const result = await discoverVertexModelsWithApiKey({
+    apiKey: "authorization-key",
+    fetchImpl: async (url, init) => {
+      urls.push(url);
+      assert.equal(new Headers(init.headers).get("x-goog-api-key"), "authorization-key");
+      assert.ok(!url.includes("authorization-key"));
+
+      return Response.json(
+        {
+          error: {
+            details: [
+              {
+                "@type": "type.googleapis.com/google.rpc.ErrorInfo",
+                metadata: { consumer: "projects/project-from-key" },
+              },
+            ],
+          },
+        },
+        { status: 403 }
+      );
+    },
+  });
+
+  assert.deepEqual(result.models, []);
+  assert.equal(result.projectId, "project-from-key");
+  assert.equal(urls.length, 1);
+  assert.ok(urls[0].includes("generativelanguage.googleapis.com"));
+  assert.ok(!urls[0].includes("/publishers/"));
+});
diff --git a/tests/unit/vertex-xai-models.test.ts b/tests/unit/vertex-xai-models.test.ts
new file mode 100644
index 00000000000..f6cedf80bcb
--- /dev/null
+++ b/tests/unit/vertex-xai-models.test.ts
@@ -0,0 +1,142 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import { getModelTargetFormat } from "../../open-sse/config/providerModels.ts";
+import { getRegistryEntry } from "../../open-sse/config/providerRegistry.ts";
+import {
+  getVertexModelTargetFormat,
+  getVertexModelTransport,
+  isVertexXaiModel,
+  normalizeVertexModelId,
+} from "../../open-sse/config/vertexModels.ts";
+import { parseVertexXaiModels } from "../../src/lib/providerModels/vertexXaiModelsParser.ts";
+
+test("normalizeVertexModelId accepts Model Garden resource names and request names", () => {
+  const cases = [
+    "grok-4.6",
+    "xai/grok-4.6",
+    "xai/models/grok-4.6",
+    "publishers/xai/models/grok-4.6",
+    "projects/demo/locations/global/publishers/xai/models/grok-4.6",
+    "vertex/publishers/xai/models/grok-4.6",
+    "vp/xai/grok-4.6",
+  ];
+
+  for (const modelId of cases) {
+    assert.equal(normalizeVertexModelId(modelId), "xai/grok-4.6", modelId);
+    assert.equal(isVertexXaiModel(modelId), true, modelId);
+  }
+
+  assert.equal(normalizeVertexModelId("gemini-3.1-pro-preview"), "gemini-3.1-pro-preview");
+  assert.equal(isVertexXaiModel("gemini-3.1-pro-preview"), false);
+});
+
+test("parseVertexXaiModels maps the Model Garden publisher envelope to routable models", () => {
+  const models = parseVertexXaiModels({
+    publisherModels: [
+      {
+        name: "publishers/xai/models/grok-4.6",
+        displayName: "Grok 4.6",
+        description: "xAI partner model",
+      },
+      {
+        name: "projects/demo/locations/global/publishers/xai/models/grok-4.20-reasoning",
+      },
+    ],
+  });
+
+  assert.deepEqual(models, [
+    {
+      id: "xai/grok-4.6",
+      name: "Grok 4.6",
+      supportedEndpoints: ["chat"],
+      targetFormat: "openai",
+      description: "xAI partner model",
+      owned_by: "xai",
+    },
+    {
+      id: "xai/grok-4.20-reasoning",
+      name: "xai/grok-4.20-reasoning",
+      supportedEndpoints: ["chat"],
+      targetFormat: "openai",
+      owned_by: "xai",
+    },
+  ]);
+});
+
+test("Vertex xAI models always resolve to the OpenAI wire format", () => {
+  const ids = ["grok-4.6", "xai/grok-4.6", "xai/models/grok-4.6", "publishers/xai/models/grok-4.6"];
+
+  for (const provider of ["vertex", "vertex-partner", "vp"]) {
+    for (const modelId of ids) {
+      assert.equal(getModelTargetFormat(provider, modelId), "openai", `${provider}/${modelId}`);
+    }
+  }
+});
+
+test("all curated and generic Vertex partner IDs resolve away from Gemini format", () => {
+  const openAiIds = [
+    "deepseek-ai/deepseek-v3.2-maas",
+    "qwen/qwen3-next-80b-a3b-instruct-maas",
+    "zai-org/glm-5-maas",
+    "mistralai/mistral-medium-3",
+    "publishers/meta/models/llama-4-maverick-17b-128e-instruct-maas",
+    "publishers/future-vendor/models/future-chat-maas",
+  ];
+
+  for (const provider of ["vertex", "vertex-partner", "vp"]) {
+    for (const modelId of openAiIds) {
+      assert.equal(getModelTargetFormat(provider, modelId), "openai", `${provider}/${modelId}`);
+    }
+    assert.equal(
+      getModelTargetFormat(provider, "publishers/anthropic/models/claude-sonnet-5"),
+      "claude",
+      provider
+    );
+  }
+});
+
+test("Vertex routing covers native partner APIs and arbitrary OpenAI-compatible MaaS publishers", () => {
+  assert.equal(
+    normalizeVertexModelId("publishers/anthropic/models/claude-sonnet-5"),
+    "claude-sonnet-5"
+  );
+  assert.equal(
+    normalizeVertexModelId("publishers/mistralai/models/mistral-medium-3"),
+    "mistral-medium-3"
+  );
+  assert.equal(
+    normalizeVertexModelId("publishers/meta/models/llama-4-maverick-17b-128e-instruct-maas"),
+    "meta/llama-4-maverick-17b-128e-instruct-maas"
+  );
+  assert.equal(
+    normalizeVertexModelId("publishers/deepseek-ai/models/deepseek-v3.2-maas"),
+    "deepseek-ai/deepseek-v3.2-maas"
+  );
+  assert.equal(normalizeVertexModelId("DeepSeek-V4-Pro"), "deepseek-ai/DeepSeek-V4-Pro");
+  assert.equal(normalizeVertexModelId("Qwen3.6-35B-A3B"), "qwen/Qwen3.6-35B-A3B");
+  assert.equal(normalizeVertexModelId("GLM-5.1-FP8"), "zai-org/GLM-5.1-FP8");
+
+  assert.equal(getVertexModelTransport("gemini-3.1-pro-preview"), "gemini");
+  assert.equal(getVertexModelTransport("anthropic/claude-sonnet-5"), "anthropic");
+  assert.equal(getVertexModelTransport("mistralai/mistral-medium-3"), "mistral");
+  assert.equal(getVertexModelTransport("future-publisher/future-chat-maas"), "openai");
+  assert.equal(getVertexModelTargetFormat("anthropic/claude-sonnet-5"), "claude");
+  assert.equal(getVertexModelTargetFormat("mistralai/mistral-medium-3"), "openai");
+  assert.equal(getVertexModelTargetFormat("future-publisher/future-chat-maas"), "openai");
+  assert.equal(getVertexModelTargetFormat("gemini-3.1-pro-preview"), null);
+});
+
+test("Vertex registries contain the documented xAI MaaS models as partial live catalogs", () => {
+  for (const provider of ["vertex", "vertex-partner"]) {
+    const entry = getRegistryEntry(provider);
+    assert.ok(entry, provider);
+    assert.equal(entry.liveCatalogAuthoritative, false, provider);
+
+    const grok = entry.models.find((model) => model.id === "xai/grok-4.6");
+    assert.ok(grok, `${provider} must expose grok-4.6`);
+    assert.equal(grok.targetFormat, "openai");
+    assert.equal(grok.supportsVision, true);
+    assert.equal(grok.contextLength, 524288);
+  }
+});