diff --git a/packages/opencode/src/session/llm.ts b/packages/opencode/src/session/llm.ts
index a99f8acff20c..6be09c145775 100644
--- a/packages/opencode/src/session/llm.ts
+++ b/packages/opencode/src/session/llm.ts
@@ -369,7 +369,7 @@ const live: Layer.Layer<
 
             // Adapter seam: both runtimes expose the same LLMEvent stream. Native
             // already returns one; AI SDK streams are converted here.
-            const state = LLMAISDK.adapterState()
+            const state = LLMAISDK.adapterState(input.model.providerID)
             return Stream.fromAsyncIterable(result.result.fullStream, (e) =>
               e instanceof Error ? e : new Error(String(e)),
             ).pipe(
diff --git a/packages/opencode/src/session/llm/ai-sdk.ts b/packages/opencode/src/session/llm/ai-sdk.ts
index 13d427aab62c..77bb9343f4e1 100644
--- a/packages/opencode/src/session/llm/ai-sdk.ts
+++ b/packages/opencode/src/session/llm/ai-sdk.ts
@@ -7,8 +7,9 @@ import { ProviderError } from "@/provider/error"
 type Result = Awaited<ReturnType<typeof streamText>>
 type AISDKEvent = Result["fullStream"] extends AsyncIterable<infer T> ? T : never
 
-export function adapterState() {
+export function adapterState(providerID?: string) {
   return {
+    providerID,
     step: 0,
     text: 0,
     reasoning: 0,
@@ -16,6 +17,7 @@ export function adapterState() {
     currentReasoningID: undefined as string | undefined,
     toolNames: {} as Record<string, string>,
     copilotTotalNanoAiu: undefined as number | undefined,
+    hasOutput: false,
   }
 }
 
@@ -88,6 +90,16 @@ export function toLLMEvents(
     case "finish-step":
       if (event.rawFinishReason === "network_error")
         return Effect.fail(new ProviderError.ResponseStreamError("Provider finish_reason: network_error"))
+      if (
+        event.finishReason === "stop" &&
+        state.providerID === "openrouter" &&
+        state.step === 0 &&
+        !state.hasOutput &&
+        !event.usage.inputTokens &&
+        !event.usage.outputTokens &&
+        !event.usage.totalTokens
+      )
+        return Effect.fail(new ProviderError.ResponseStreamError("Provider returned an empty response"))
       return Effect.sync(() => {
         const original = providerMetadata(event.providerMetadata)
         const metadata =
@@ -101,6 +113,7 @@ export function toLLMEvents(
                 },
               }
         state.copilotTotalNanoAiu = undefined
+        state.hasOutput = false
         return [
           LLMEvent.stepFinish({
             index: state.step++,
@@ -122,7 +135,7 @@ export function toLLMEvents(
         ]
         // Reset so the adapter can be reused for a follow-up stream without leaking
         // counters or block IDs. adapterState() is the single source of truth for shape.
-        Object.assign(state, adapterState())
+        Object.assign(state, adapterState(state.providerID))
         return events
       })
 
@@ -138,13 +151,16 @@ export function toLLMEvents(
       })
 
     case "text-delta":
-      return Effect.succeed([
-        LLMEvent.textDelta({
-          id: currentTextID(state, event.id),
-          text: event.text,
-          providerMetadata: providerMetadata(event.providerMetadata),
-        }),
-      ])
+      return Effect.sync(() => {
+        state.hasOutput ||= event.text.length > 0
+        return [
+          LLMEvent.textDelta({
+            id: currentTextID(state, event.id),
+            text: event.text,
+            providerMetadata: providerMetadata(event.providerMetadata),
+          }),
+        ]
+      })
 
     case "text-end":
       return Effect.sync(() => {
@@ -170,13 +186,16 @@ export function toLLMEvents(
       })
 
     case "reasoning-delta":
-      return Effect.succeed([
-        LLMEvent.reasoningDelta({
-          id: currentReasoningID(state, event.id),
-          text: event.text,
-          providerMetadata: providerMetadata(event.providerMetadata),
-        }),
-      ])
+      return Effect.sync(() => {
+        state.hasOutput ||= event.text.length > 0
+        return [
+          LLMEvent.reasoningDelta({
+            id: currentReasoningID(state, event.id),
+            text: event.text,
+            providerMetadata: providerMetadata(event.providerMetadata),
+          }),
+        ]
+      })
 
     case "reasoning-end":
       return Effect.sync(() => {
@@ -222,6 +241,7 @@ export function toLLMEvents(
 
     case "tool-call":
       return Effect.sync(() => {
+        state.hasOutput = true
         state.toolNames[event.toolCallId] = event.toolName
         return [
           LLMEvent.toolCall({
diff --git a/packages/opencode/test/fixtures/openrouter-ox-duplicate-stop.json b/packages/opencode/test/fixtures/openrouter-ox-duplicate-stop.json
new file mode 100644
index 000000000000..3a6d110d8a6c
--- /dev/null
+++ b/packages/opencode/test/fixtures/openrouter-ox-duplicate-stop.json
@@ -0,0 +1,64 @@
+{
+  "chunks": [
+    {
+      "id": "chatcmpl-ox",
+      "model": "stealth/ox-alpha",
+      "provider": "Stealth",
+      "choices": [
+        {
+          "index": 0,
+          "finish_reason": null,
+          "delta": {
+            "role": "assistant",
+            "content": "",
+            "reasoning": "thinking",
+            "reasoning_details": [{ "type": "reasoning.text", "text": "thinking" }]
+          }
+        }
+      ]
+    },
+    {
+      "id": "chatcmpl-ox",
+      "model": "stealth/ox-alpha",
+      "provider": "Stealth",
+      "choices": [
+        {
+          "index": 0,
+          "finish_reason": null,
+          "delta": { "role": "assistant", "content": "OX_STREAM_OK" }
+        }
+      ]
+    },
+    {
+      "id": "chatcmpl-ox",
+      "model": "stealth/ox-alpha",
+      "provider": "Stealth",
+      "choices": [
+        {
+          "index": 0,
+          "finish_reason": "stop",
+          "delta": { "role": "assistant", "content": "", "reasoning": "" }
+        }
+      ]
+    },
+    {
+      "id": "chatcmpl-ox",
+      "model": "stealth/ox-alpha",
+      "provider": "Stealth",
+      "choices": [
+        {
+          "index": 0,
+          "finish_reason": "stop",
+          "delta": { "role": "assistant", "content": "" }
+        }
+      ],
+      "usage": {
+        "prompt_tokens": 96,
+        "completion_tokens": 52,
+        "total_tokens": 148,
+        "prompt_tokens_details": { "cached_tokens": 64 },
+        "completion_tokens_details": { "reasoning_tokens": 0 }
+      }
+    }
+  ]
+}
diff --git a/packages/opencode/test/session/llm.test.ts b/packages/opencode/test/session/llm.test.ts
index 635e697517c3..223d00059079 100644
--- a/packages/opencode/test/session/llm.test.ts
+++ b/packages/opencode/test/session/llm.test.ts
@@ -9,6 +9,7 @@ import { InstanceRef } from "../../src/effect/instance-ref"
 import { HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
 import z from "zod"
 import { LLM } from "../../src/session/llm"
+import type { LLMEvent } from "@opencode-ai/llm"
 import { LLMClient, RequestExecutor } from "@opencode-ai/llm/route"
 import { Provider } from "@/provider/provider"
 import { ProviderTransform } from "@/provider/transform"
@@ -57,6 +58,8 @@ const it = testEffect(AppNodeBuilder.build(LayerNode.group([LLM.node, Provider.n
 
 // LLM.stream returns a Stream, not an Effect, so we can't use the serviceUse proxy.
 const drain = (input: LLM.StreamInput) => LLM.Service.use((svc) => svc.stream(input).pipe(Stream.runDrain))
+const collect = (input: LLM.StreamInput) =>
+  LLM.Service.use((svc) => svc.stream(input).pipe(Stream.runCollect, Effect.map(Array.from)))
 
 // drainWith builds an isolated runtime so custom replacements fully own LLM and
 // its transitive deps.
@@ -362,6 +365,7 @@ describe("session.llm.ai-sdk adapter", () => {
     // telemetry distinguishes "missing" from "zero," so emitting an empty object causes
     // false positives ("usage was tracked, just empty") instead of correct nulls.
     const events = await adapt([
+      uncheckedAdapterEvent({ type: "text-delta", text: "ok" }),
       {
         type: "finish-step",
         response: { id: "response-1", timestamp: new Date(0), modelId: "gpt-test" },
@@ -380,8 +384,8 @@ describe("session.llm.ai-sdk adapter", () => {
       },
     ])
 
-    expect(events).toHaveLength(1)
-    const stepFinish = events[0]
+    expect(events).toHaveLength(2)
+    const stepFinish = events[1]
     if (stepFinish.type !== "step-finish") throw new Error("expected step-finish")
     expect(stepFinish.usage).toBeUndefined()
   })
@@ -897,6 +901,73 @@ describe("session.llm.stream", () => {
     },
   )
 
+  const openrouterFixture = { providerID: "openrouter", modelID: "inclusionai/ling-2.6-1t" }
+  it.instance(
+    "preserves OX Alpha text and usage across duplicate stop chunks",
+    () =>
+      Effect.gen(function* () {
+        const fixture = loadFixture(openrouterFixture.providerID, openrouterFixture.modelID)
+        const stream = (yield* Effect.promise(() =>
+          Bun.file(path.join(import.meta.dir, "../fixtures/openrouter-ox-duplicate-stop.json")).json(),
+        )) as { chunks: unknown[] }
+        const request = waitRequest("/chat/completions", createEventResponse(stream.chunks, true))
+        const resolved = yield* Provider.use.getModel(
+          ProviderV2.ID.make(openrouterFixture.providerID),
+          ModelV2.ID.make(fixture.model.id),
+        )
+        const sessionID = SessionID.make("session-test-ox-duplicate-stop")
+        const agent = {
+          name: "test",
+          mode: "primary",
+          options: {},
+          permission: [{ permission: "*", pattern: "*", action: "allow" }],
+        } satisfies Agent.Info
+        const user = {
+          id: MessageID.make("msg_user-ox-duplicate-stop"),
+          sessionID,
+          role: "user",
+          time: { created: Date.now() },
+          agent: agent.name,
+          model: { providerID: ProviderV2.ID.make(openrouterFixture.providerID), modelID: resolved.id },
+        } satisfies SessionV1.User
+
+        // oxlint-disable-next-line typescript-eslint/no-unsafe-type-assertion -- LLM.Service.use erases the stream item type
+        const events = (yield* collect({
+          user,
+          sessionID,
+          model: resolved,
+          agent,
+          system: ["You are a helpful assistant."],
+          messages: [{ role: "user", content: "Reply with exactly OX_STREAM_OK" }],
+          tools: {},
+        })) as LLMEvent[]
+        yield* Effect.promise(() => request)
+
+        expect(
+          events
+            .filter((event) => event.type === "text-delta")
+            .map((event) => event.text)
+            .join(""),
+        ).toBe("OX_STREAM_OK")
+        expect(events.find((event) => event.type === "step-finish")?.usage).toMatchObject({
+          inputTokens: 96,
+          outputTokens: 52,
+          totalTokens: 148,
+          cacheReadInputTokens: 64,
+        })
+      }),
+    {
+      config: () => ({
+        enabled_providers: [openrouterFixture.providerID],
+        provider: {
+          [openrouterFixture.providerID]: {
+            options: { apiKey: "test-key", baseURL: `${state.server!.url.origin}/v1` },
+          },
+        },
+      }),
+    },
+  )
+
   const cerebrasFixture = { providerID: "cerebras", modelID: "gpt-oss-120b" }
   it.instance(
     "replays Cerebras assistant reasoning using the provider-supported field",
diff --git a/packages/opencode/test/session/processor-effect.test.ts b/packages/opencode/test/session/processor-effect.test.ts
index c67f82d9c71b..36b6ff71d8f3 100644
--- a/packages/opencode/test/session/processor-effect.test.ts
+++ b/packages/opencode/test/session/processor-effect.test.ts
@@ -41,6 +41,11 @@ const ref = {
   modelID: ModelV2.ID.make("test-model"),
 }
 
+const openrouterRef = {
+  providerID: ProviderV2.ID.make("openrouter"),
+  modelID: ModelV2.ID.make("test-model"),
+}
+
 const cfg = {
   provider: {
     test: {
@@ -86,6 +91,22 @@ function providerCfg(url: string) {
   }
 }
 
+function openrouterCfg(url: string) {
+  return {
+    provider: {
+      openrouter: {
+        ...cfg.provider.test,
+        id: "openrouter",
+        npm: "@openrouter/ai-sdk-provider",
+        options: {
+          ...cfg.provider.test.options,
+          baseURL: url,
+        },
+      },
+    },
+  }
+}
+
 function agent(): Agent.Info {
   return {
     name: "build",
@@ -708,6 +729,69 @@ it.live("session.processor effect tests retry network_error finish reasons", ()
   ),
 )
 
+it.live("session.processor effect tests retry empty successful streams", () =>
+  provideTmpdirServer(
+    ({ dir, llm }) =>
+      Effect.gen(function* () {
+        const { processors, session, provider } = yield* boot()
+
+        yield* llm.push(
+          raw({
+            chunks: [
+              {
+                id: "chatcmpl-empty-stop",
+                object: "chat.completion.chunk",
+                choices: [{ index: 0, delta: { role: "assistant", content: "" }, finish_reason: "stop" }],
+              },
+              {
+                id: "chatcmpl-empty-stop",
+                object: "chat.completion.chunk",
+                choices: [{ index: 0, delta: { role: "assistant", content: "" }, finish_reason: "stop" }],
+                usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 },
+              },
+            ],
+          }),
+        )
+        yield* llm.text("after empty retry")
+
+        const chat = yield* session.create({})
+        const parent = yield* user(chat.id, "retry empty response")
+        const msg = yield* assistant(chat.id, parent.id, path.resolve(dir))
+        const mdl = yield* provider.getModel(openrouterRef.providerID, openrouterRef.modelID)
+        const handle = yield* processors.create({
+          assistantMessage: msg,
+          sessionID: chat.id,
+          model: mdl,
+        })
+
+        const value = yield* handle.process({
+          user: {
+            id: parent.id,
+            sessionID: chat.id,
+            role: "user",
+            time: parent.time,
+            agent: parent.agent,
+            model: { providerID: openrouterRef.providerID, modelID: openrouterRef.modelID },
+          } satisfies SessionV1.User,
+          sessionID: chat.id,
+          model: mdl,
+          agent: agent(),
+          system: [],
+          messages: [{ role: "user", content: "retry empty response" }],
+          tools: {},
+        })
+
+        const parts = yield* MessageV2.parts(msg.id)
+
+        expect(value).toBe("continue")
+        expect(yield* llm.calls).toBe(2)
+        expect(parts.some((part) => part.type === "text" && part.text === "after empty retry")).toBe(true)
+        expect(handle.message.error).toBeUndefined()
+      }),
+    { config: (url) => openrouterCfg(url) },
+  ),
+)
+
 it.live("session.processor effect tests publish retry status updates", () =>
   provideTmpdirServer(
     ({ dir, llm }) =>