diff --git a/changelog.d/fixes/16579-opencode-go-context-limits.md b/changelog.d/fixes/16579-opencode-go-context-limits.md
new file mode 100644
index 000000000000..5d606e3064f4
--- /dev/null
+++ b/changelog.d/fixes/16579-opencode-go-context-limits.md
@@ -0,0 +1 @@
+- Correct the context and output limits for twelve verified OpenCode Go models so imported models no longer fall back to a 200K context window. Preserve the narrower GPT-6 Luna and Space Bunny input limits, operator overrides, synced metadata, and documented request formats.
diff --git a/open-sse/config/providers/registry/opencode/go/index.ts b/open-sse/config/providers/registry/opencode/go/index.ts
index 303a4b358914..7c531535c55c 100644
--- a/open-sse/config/providers/registry/opencode/go/index.ts
+++ b/open-sse/config/providers/registry/opencode/go/index.ts
@@ -110,6 +110,8 @@ export const opencode_goProvider: RegistryEntry = {
     {
       id: "qwen3.7-plus",
       name: "Qwen3.7 Plus",
+      contextLength: 1000000,
+      maxOutputTokens: 65536,
       targetFormat: "claude",
       supportsVision: false,
       supportsReasoning: true,
@@ -372,6 +374,8 @@ export const opencode_goProvider: RegistryEntry = {
     {
       id: "grok-4.6",
       name: "Grok 4.6",
+      contextLength: 500000,
+      maxOutputTokens: 500000,
       supportsReasoning: true,
       supportedThinkingEfforts: ["low", "medium", "high", "xhigh"],
       targetFormat: "openai-responses",
@@ -417,11 +421,74 @@ export const opencode_goProvider: RegistryEntry = {
     {
       id: "gpt-6-luna",
       name: "GPT-6 Luna",
+      contextLength: 1050000,
+      maxInputTokens: 922000,
       supportsReasoning: true,
       supportedThinkingEfforts: ["none", "low", "medium", "high", "xhigh", "max"],
       targetFormat: "openai-responses",
       maxOutputTokens: 128000,
     },
+    // #16579: exact Go limits from anomalyco/models.dev at
+    // 26d4dc7b66c93470f4e5337708e794b62a589620, including base_model inheritance.
+    // Keep these provider-scoped so synced data and operator overrides retain precedence.
+    {
+      id: "deepseek-v4.1-flash",
+      name: "DeepSeek V4.1 Flash",
+      contextLength: 1000000,
+      maxOutputTokens: 384000,
+    },
+    {
+      id: "claude-haiku-5-5",
+      name: "Claude Haiku 5.5",
+      contextLength: 1000000,
+      maxOutputTokens: 128000,
+      targetFormat: "claude",
+    },
+    {
+      id: "grok-4.7",
+      name: "Grok 4.7",
+      contextLength: 500000,
+      maxOutputTokens: 500000,
+      targetFormat: "openai-responses",
+    },
+    {
+      id: "hy4-preview",
+      name: "HY4 Preview",
+      contextLength: 1024000,
+      maxOutputTokens: 64000,
+    },
+    {
+      id: "longcat-2.0",
+      name: "LongCat 2.0",
+      contextLength: 1000000,
+      maxOutputTokens: 131072,
+    },
+    {
+      id: "mimo-v2.6-flash",
+      name: "MiMo V2.6 Flash",
+      contextLength: 1048576,
+      maxOutputTokens: 131072,
+    },
+    {
+      id: "mimo-v2.6-pro",
+      name: "MiMo V2.6 Pro",
+      contextLength: 1048576,
+      maxOutputTokens: 131072,
+    },
+    {
+      id: "qwen3.8-flash",
+      name: "Qwen3.8 Flash",
+      contextLength: 1000000,
+      maxOutputTokens: 131072,
+      targetFormat: "claude",
+    },
+    {
+      id: "space-bunny",
+      name: "Space Bunny",
+      contextLength: 1048576,
+      maxInputTokens: 524288,
+      maxOutputTokens: 524288,
+    },
     // Console Go free GLM-tier model (live-verified 2026-08-23): the upstream
     // rejects every reasoning_effort outside {low, high, max} whenever tools
     // are present — "[1210] This model always engages in thinking and cannot
diff --git a/tests/unit/issue-16579-opencode-go-context-limits.test.ts b/tests/unit/issue-16579-opencode-go-context-limits.test.ts
new file mode 100644
index 000000000000..70cb6401581c
--- /dev/null
+++ b/tests/unit/issue-16579-opencode-go-context-limits.test.ts
@@ -0,0 +1,158 @@
+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";
+import type { ModelCapabilityEntry } from "../../src/lib/modelsDevSync/transform.ts";
+
+const dataDir = fs.mkdtempSync(path.join(os.tmpdir(), "omni-go-limits-"));
+process.env.DATA_DIR = dataDir;
+const savedContextEnv = new Map(
+  ["CONTEXT_LENGTH_OPENCODE_GO", "CONTEXT_LENGTH_DEFAULT"].map((name) => [name, process.env[name]])
+);
+for (const name of savedContextEnv.keys()) delete process.env[name];
+const core = await import("../../src/lib/db/core.ts");
+const { getResolvedModelCapabilities, createModelCapabilityResolutionSnapshot } =
+  await import("../../src/lib/modelCapabilities.ts");
+const { enrichCatalogModelEntry } = await import("../../src/lib/modelMetadataRegistry.ts");
+const { getModelTargetFormat } = await import("../../open-sse/config/providerModels.ts");
+const { getTokenLimit } = await import("../../open-sse/services/contextManager.ts");
+test.after(() => {
+  for (const [name, value] of savedContextEnv) {
+    if (value === undefined) delete process.env[name];
+    else process.env[name] = value;
+  }
+  core.resetDbInstance();
+  fs.rmSync(dataDir, { recursive: true, force: true });
+});
+
+// Exact provider/base-model TOMLs at anomalyco/models.dev commit
+// 26d4dc7b66c93470f4e5337708e794b62a589620; never infer an integer from "1M".
+const limits: readonly (readonly [string, number, number])[] = [
+  ["deepseek-v4.1-flash", 1000000, 384000],
+  ["claude-haiku-5-5", 1000000, 128000],
+  ["gpt-6-luna", 1050000, 128000],
+  ["grok-4.6", 500000, 500000],
+  ["grok-4.7", 500000, 500000],
+  ["hy4-preview", 1024000, 64000],
+  ["longcat-2.0", 1000000, 131072],
+  ["mimo-v2.6-flash", 1048576, 131072],
+  ["mimo-v2.6-pro", 1048576, 131072],
+  ["qwen3.7-plus", 1000000, 65536],
+  ["qwen3.8-flash", 1000000, 131072],
+  ["space-bunny", 1048576, 524288],
+];
+
+for (const [model, context, output] of limits) {
+  test(`#16579 ${model} exposes exact Go limits without models.dev sync`, () => {
+    const snapshot = createModelCapabilityResolutionSnapshot();
+    assert.equal(Object.keys(snapshot.synced).length, 0);
+    const caps = getResolvedModelCapabilities(
+      { provider: "opencode-go", model },
+      undefined,
+      snapshot
+    );
+    assert.equal(caps.contextWindow, context);
+    assert.equal(getTokenLimit("opencode-go", model, snapshot), context);
+    assert.equal(caps.maxOutputTokens, output);
+    const imported: Record<string, unknown> = {
+      id: `opencode-go/${model}`,
+      root: model,
+      owned_by: "opencode-go",
+      name: model,
+    };
+    const row = enrichCatalogModelEntry(imported, undefined, {
+      modelsDevPricing: {},
+      userPricing: {},
+      capabilityResolutionSnapshot: snapshot,
+    });
+    assert.equal(row.context_length, context);
+    assert.equal(row.max_output_tokens, output);
+  });
+}
+
+test("#16579 explicit input ceilings stay narrower than the full context", () => {
+  const snapshot = createModelCapabilityResolutionSnapshot();
+  for (const [model, expected] of [
+    ["gpt-6-luna", 922000],
+    ["space-bunny", 524288],
+  ] as const) {
+    const caps = getResolvedModelCapabilities(
+      { provider: "opencode-go", model },
+      undefined,
+      snapshot
+    );
+    assert.equal(caps.maxInputTokens, expected);
+  }
+});
+
+test("#16579 synced and persisted context/output overrides retain precedence", () => {
+  const row: ModelCapabilityEntry = {
+    tool_call: null,
+    reasoning: null,
+    attachment: null,
+    structured_output: null,
+    temperature: null,
+    modalities_input: "[]",
+    modalities_output: "[]",
+    knowledge_cutoff: null,
+    release_date: null,
+    last_updated: null,
+    status: null,
+    family: null,
+    open_weights: null,
+    limit_context: 345678,
+    limit_input: 300000,
+    limit_output: 12000,
+    interleaved_field: null,
+  };
+  const snapshot = {
+    ...createModelCapabilityResolutionSnapshot(),
+    synced: { "opencode-go": { "grok-4.7": row } },
+  };
+  const caps = getResolvedModelCapabilities("opencode-go/grok-4.7", undefined, snapshot);
+  assert.equal(caps.contextWindow, 345678);
+  assert.equal(caps.maxOutputTokens, 12000);
+  assert.equal(caps.maxInputTokens, 300000);
+  const operatorSnapshot = {
+    ...snapshot,
+    contextOverrides: new Map([["opencode-go", new Map([["grok-4.7", 400000]])]]),
+    maxTokenOverrides: new Map([["opencode-go", new Map([["grok-4.7", 16000]])]]),
+    maxInputTokenOverrides: new Map([["opencode-go", new Map([["grok-4.7", 200000]])]]),
+  };
+  const overridden = getResolvedModelCapabilities(
+    "opencode-go/grok-4.7",
+    undefined,
+    operatorSnapshot
+  );
+  assert.equal(overridden.contextWindow, 400000);
+  assert.equal(overridden.maxOutputTokens, 16000);
+  assert.equal(overridden.maxInputTokens, 200000);
+});
+
+test("#16579 catalog rows preserve the documented Go request formats", () => {
+  for (const [model, format] of [
+    ["gpt-6-luna", "openai-responses"],
+    ["grok-4.6", "openai-responses"],
+    ["grok-4.7", "openai-responses"],
+    ["claude-haiku-5-5", "claude"],
+    ["qwen3.7-plus", "claude"],
+    ["qwen3.8-flash", "claude"],
+  ] as const)
+    assert.equal(getModelTargetFormat("opencode-go", model), format, model);
+});
+
+test("#16579 undocumented models do not inherit a guessed family window", () => {
+  const snapshot = createModelCapabilityResolutionSnapshot();
+  const caps = getResolvedModelCapabilities("opencode-go/omen-alpha", undefined, snapshot);
+  assert.equal(caps.contextWindow, null);
+});
+
+test("#16579 explicit environment limits still override catalog context", () => {
+  process.env.CONTEXT_LENGTH_OPENCODE_GO = "123456";
+  try {
+    assert.equal(getTokenLimit("opencode-go", "grok-4.7"), 123456);
+  } finally {
+    delete process.env.CONTEXT_LENGTH_OPENCODE_GO;
+  }
+});