diff --git a/packages/opencode/src/session/llm/ai-sdk.ts b/packages/opencode/src/session/llm/ai-sdk.ts
index 8db8985d7b09..b4a17433788d 100644
--- a/packages/opencode/src/session/llm/ai-sdk.ts
+++ b/packages/opencode/src/session/llm/ai-sdk.ts
@@ -13,6 +13,8 @@ export function adapterState() {
     reasoning: 0,
     currentTextID: undefined as string | undefined,
     currentReasoningID: undefined as string | undefined,
+    startedText: {} as Record<string, boolean>,
+    startedReasoning: {} as Record<string, boolean>,
     toolNames: {} as Record<string, string>,
     copilotTotalNanoAiu: undefined as number | undefined,
   }
@@ -73,6 +75,18 @@ function currentReasoningID(state: ReturnType<typeof adapterState>, id: string |
   return state.currentReasoningID
 }
 
+// AI SDK reports missing stream starts in-band, then continues streaming.
+function isOrphanStreamStateError(error: unknown) {
+  const message = errorMessage(error).trim()
+  return message.endsWith(" not found") && (message.startsWith("reasoning part ") || message.startsWith("text part "))
+}
+
+function synthesizeStart(started: Record<string, boolean>, id: string, make: () => LLMEvent): LLMEvent[] {
+  if (started[id]) return []
+  started[id] = true
+  return [make()]
+}
+
 export function toLLMEvents(
   state: ReturnType<typeof adapterState>,
   event: AISDKEvent,
@@ -126,6 +140,7 @@ export function toLLMEvents(
     case "text-start":
       return Effect.sync(() => {
         state.currentTextID = currentTextID(state, event.id)
+        state.startedText[state.currentTextID] = true
         return [
           LLMEvent.textStart({
             id: state.currentTextID,
@@ -135,19 +150,28 @@ 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(() => {
+        const id = currentTextID(state, event.id)
+        return [
+          ...synthesizeStart(state.startedText, id, () =>
+            LLMEvent.textStart({ id, providerMetadata: providerMetadata(event.providerMetadata) }),
+          ),
+          LLMEvent.textDelta({
+            id,
+            text: event.text,
+            providerMetadata: providerMetadata(event.providerMetadata),
+          }),
+        ]
+      })
 
     case "text-end":
       return Effect.sync(() => {
         const id = currentTextID(state, event.id)
         state.currentTextID = undefined
         return [
+          ...synthesizeStart(state.startedText, id, () =>
+            LLMEvent.textStart({ id, providerMetadata: providerMetadata(event.providerMetadata) }),
+          ),
           LLMEvent.textEnd({
             id,
             providerMetadata: providerMetadata(event.providerMetadata),
@@ -158,6 +182,7 @@ export function toLLMEvents(
     case "reasoning-start":
       return Effect.sync(() => {
         state.currentReasoningID = currentReasoningID(state, event.id)
+        state.startedReasoning[state.currentReasoningID] = true
         return [
           LLMEvent.reasoningStart({
             id: state.currentReasoningID,
@@ -167,19 +192,28 @@ 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(() => {
+        const id = currentReasoningID(state, event.id)
+        return [
+          ...synthesizeStart(state.startedReasoning, id, () =>
+            LLMEvent.reasoningStart({ id, providerMetadata: providerMetadata(event.providerMetadata) }),
+          ),
+          LLMEvent.reasoningDelta({
+            id,
+            text: event.text,
+            providerMetadata: providerMetadata(event.providerMetadata),
+          }),
+        ]
+      })
 
     case "reasoning-end":
       return Effect.sync(() => {
         const id = currentReasoningID(state, event.id)
         state.currentReasoningID = undefined
         return [
+          ...synthesizeStart(state.startedReasoning, id, () =>
+            LLMEvent.reasoningStart({ id, providerMetadata: providerMetadata(event.providerMetadata) }),
+          ),
           LLMEvent.reasoningEnd({
             id,
             providerMetadata: providerMetadata(event.providerMetadata),
@@ -262,6 +296,10 @@ export function toLLMEvents(
       })
 
     case "error":
+      if (isOrphanStreamStateError(event.error))
+        return Effect.logDebug("dropping orphan reasoning/text stream part", {
+          detail: errorMessage(event.error),
+        }).pipe(Effect.as<LLMEvent[]>([]))
       return Effect.fail(event.error)
 
     case "abort":
diff --git a/packages/opencode/test/session/llm.test.ts b/packages/opencode/test/session/llm.test.ts
index 3bfc722e2bec..d4999b6c33ef 100644
--- a/packages/opencode/test/session/llm.test.ts
+++ b/packages/opencode/test/session/llm.test.ts
@@ -313,13 +313,42 @@ describe("session.llm.ai-sdk adapter", () => {
     ])
 
     expect(events).toMatchObject([
+      { type: "text-start", id: "text-0" },
       { type: "text-delta", id: "text-0", text: "implicit text" },
       { type: "text-end", id: "text-0" },
+      { type: "reasoning-start", id: "reasoning-0" },
       { type: "reasoning-delta", id: "reasoning-0", text: "implicit reasoning" },
       { type: "reasoning-end", id: "reasoning-0" },
     ])
   })
 
+  test("repairs orphan stream parts without hiding provider errors", async () => {
+    const metadata = { openai: { itemID: "rs_abc" } }
+    const events = await adapt([
+      uncheckedAdapterEvent({
+        type: "reasoning-delta",
+        id: "rs_abc:0",
+        text: "thinking",
+        providerMetadata: metadata,
+      }),
+      uncheckedAdapterEvent({ type: "reasoning-end", id: "rs_abc:0" }),
+      uncheckedAdapterEvent({ type: "error", error: "reasoning part rs_abc:0 not found" }),
+    ])
+
+    expect(events).toMatchObject([
+      { type: "reasoning-start", id: "rs_abc:0", providerMetadata: metadata },
+      { type: "reasoning-delta", id: "rs_abc:0", text: "thinking", providerMetadata: metadata },
+      { type: "reasoning-end", id: "rs_abc:0" },
+    ])
+    const failure = await Effect.runPromiseExit(
+      LLMAISDK.toLLMEvents(
+        LLMAISDK.adapterState(),
+        uncheckedAdapterEvent({ type: "error", error: "rate limit exceeded" }),
+      ),
+    )
+    expect(Exit.isFailure(failure)).toBe(true)
+  })
+
   test("explicitly ignores non-session-visible AI SDK chunks", async () => {
     expect(
       await adapt([
@@ -441,8 +470,10 @@ describe("session.llm.ai-sdk adapter", () => {
 
     expect(secondStream).toMatchObject([
       { type: "step-start", index: 0 },
+      { type: "text-start", id: "text-0" },
       { type: "text-delta", id: "text-0", text: "second" },
       { type: "text-end", id: "text-0" },
+      { type: "reasoning-start", id: "reasoning-0" },
       { type: "reasoning-delta", id: "reasoning-0", text: "second reasoning" },
       { type: "reasoning-end", id: "reasoning-0" },
     ])