ai 7.0.34 → 7.0.35

package/dist/internal/index.d.ts

@@ -421,13 +421,15 @@ type LanguageModelCallOptions = {
  * - A number representing milliseconds
  * - An object with `totalMs` property for the total timeout in milliseconds
  * - An object with `stepMs` property for the timeout of each step in milliseconds
- * - An object with `chunkMs` property for the timeout between stream chunks (streaming only)
+ * - An object with `firstChunkMs` property for the timeout until the first content chunk of each step (streaming only)
+ * - An object with `chunkMs` property for the timeout between content chunks (streaming only)
  * - An object with `toolMs` property for the default timeout for all tool executions
  * - An object with `tools` property for per-tool timeout overrides using `{toolName}Ms` keys
  */
 type TimeoutConfiguration<TOOLS extends ToolSet> = number | {
     totalMs?: number;
     stepMs?: number;
+    firstChunkMs?: number;
     chunkMs?: number;
     toolMs?: number;
     tools?: Partial<Record<`${keyof TOOLS & string}Ms`, number>>;
@@ -1965,7 +1967,8 @@ type GenerateTextStartEvent<TOOLS extends ToolSet = ToolSet, RUNTIME_CONTEXT ext
     readonly maxRetries: number;
     /**
      * Timeout configuration for the generation.
-     * Can be a number (milliseconds) or an object with totalMs, stepMs, chunkMs, toolMs, and per-tool overrides via tools.
+     * Can be a number (milliseconds) or an object with totalMs, stepMs,
+     * firstChunkMs (streaming only), chunkMs, toolMs, and per-tool overrides via tools.
      */
     readonly timeout: TimeoutConfiguration<TOOLS> | undefined;
     /** Additional HTTP headers sent with the request. */

package/dist/internal/index.js

@@ -85,7 +85,7 @@ import {
 } from "@ai-sdk/provider-utils";
 
 // src/version.ts
-var VERSION = true ? "7.0.34" : "0.0.0-test";
+var VERSION = true ? "7.0.35" : "0.0.0-test";
 
 // src/util/download/download.ts
 var download = async ({

package/docs/03-ai-sdk-core/25-settings.mdx

@@ -142,7 +142,8 @@ You can specify the timeout either as a number (milliseconds) or as an object wi
 
 - `totalMs`: The total timeout for the entire call including all steps.
 - `stepMs`: The timeout for each individual step (LLM call). This is useful for multi-step generations where you want to limit the time spent on each step independently.
-- `chunkMs`: The timeout between stream chunks (streaming only). The call will abort if no new chunk is received within this duration. This is useful for detecting stalled streams.
+- `firstChunkMs`: The timeout until the first content-bearing output of each step (streaming only). Text deltas, reasoning deltas, tool-input deltas, generated files, and tool calls satisfy the timeout. Response metadata, stream starts, empty deltas, raw chunks, and transport activity do not satisfy or reset it.
+- `chunkMs`: The timeout between content-bearing output chunks after output has started (streaming only). Non-content chunks do not reset it. This is useful for detecting streams that stall after generation begins.
 - `toolMs`: The default timeout for all tool executions. If a tool takes longer, it aborts and returns a tool-error so the model can respond or retry.
 - `tools`: Per-tool timeout overrides using `{toolName}Ms` keys (e.g. `weatherMs`, `slowApiMs`). Takes precedence over `toolMs`. Tool names are type-checked for autocomplete.
 
@@ -195,7 +196,19 @@ const result = await generateText({
 const result = streamText({
   model: __MODEL__,
   prompt: 'Invent a new holiday and describe its traditions.',
-  timeout: { chunkMs: 5000 }, // abort if no chunk received for 5 seconds
+  timeout: { chunkMs: 5000 }, // abort if content stalls for 5 seconds
+});
+```
+
+#### Example: First-content timeout for streaming (streamText only)
+
+```ts
+const result = streamText({
+  model: __MODEL__,
+  prompt: 'Invent a new holiday and describe its traditions.',
+  timeout: {
+    firstChunkMs: 10000, // 10 seconds for first content in each step
+  },
 });
 ```
 

package/docs/03-ai-sdk-core/65-devtools.mdx

@@ -79,6 +79,12 @@ npx @ai-sdk/devtools@latest
 
 Open [http://localhost:4983](http://localhost:4983) to view your AI SDK interactions.
 
+### Choose a viewer theme
+
+The viewer uses the dark theme by default. Use the theme button in the viewer
+header to switch between dark and light themes. Your selection is stored in
+your browser and restored the next time you open the viewer at the same origin.
+
 ### Monorepo usage
 
 If you are using a monorepo setup (e.g. Turborepo, Nx), start DevTools from the same workspace where your AI SDK code runs.

package/docs/03-ai-sdk-harnesses/02-harness-agent.mdx

@@ -192,6 +192,46 @@ const result = await agent.continueStream({ session });
 Use `continueStream()` for incremental output, or `continueGenerate()` to drain
 the continued turn and return a `GenerateTextResult`.
 
+## Stop After a Harness Step
+
+Use `stopWhen` to opt into semantic step boundaries. Predicates run after real
+harness tool steps that can continue into another model step, and receive the
+completed steps from the current invocation. When a predicate matches, the
+returned result finishes while the underlying turn remains unfinished. A
+terminal text-only step instead consumes the turn's `finish` event and finishes
+naturally.
+
+```ts
+import { isStepCount } from 'ai';
+
+const steppedAgent = new HarnessAgent({
+  harness: claudeCode,
+  sandbox: createVercelSandbox({
+    runtime: 'node24',
+    ports: [4000],
+  }),
+  stopWhen: isStepCount(1),
+});
+
+const session = await steppedAgent.createSession();
+const result = await steppedAgent.generate({
+  session,
+  prompt: 'Create a short TODO.md for this repository.',
+});
+
+if (session.hasUnfinishedTurn()) {
+  const continueFrom = await session.suspendTurn();
+  await persistContinuationState({ chatId, continuationState: continueFrom });
+}
+```
+
+`stopWhen` has no default. When omitted, `HarnessAgent` continues running until
+the turn naturally finishes or pauses for host input, preserving the behavior
+of agents without step control. Pass one predicate or an array; matching any
+predicate finishes the current result slice. Resume a stopped turn with
+`createSession({ continueFrom })` and `continueStream()` or
+`continueGenerate()`.
+
 ## Prepare the Sandbox
 
 Use `sandboxConfig` to prepare the sandbox before the harness starts.
@@ -325,6 +365,8 @@ console.log(preparation.identity);
 - `sandbox`: a `HarnessV1SandboxProvider`.
 - `id`: optional stable agent identifier.
 - `instructions`: instructions applied once to a fresh session.
+- `stopWhen`: condition(s) for finishing a result slice after a completed
+  harness tool step that can continue into another model step.
 - `tools`: AI SDK tools executed by the host when the harness calls them.
 - `activeTools`: allowlist of built-in and host-executed tools the harness can
   call.

package/docs/07-reference/01-ai-sdk-core/02-stream-text.mdx

@@ -477,10 +477,10 @@ To see `streamText` in action, check out [these examples](#examples).
     },
     {
       name: 'timeout',
-      type: 'number | { totalMs?: number; stepMs?: number; chunkMs?: number; toolMs?: number; tools?: { [toolName]Ms?: number } }',
+      type: 'number | { totalMs?: number; stepMs?: number; firstChunkMs?: number; chunkMs?: number; toolMs?: number; tools?: { [toolName]Ms?: number } }',
       isOptional: true,
       description:
-        'Timeout in milliseconds. Can be specified as a number or as an object with totalMs, stepMs, chunkMs, toolMs, and/or tools properties. totalMs sets the total timeout for the entire call. stepMs sets the timeout for each individual step (LLM call). chunkMs sets the timeout between stream chunks - the call will abort if no new chunk is received within this duration. toolMs sets the default timeout for all tool executions. tools sets per-tool timeout overrides using the pattern {toolName}Ms (e.g. weatherMs, slowApiMs) that take precedence over toolMs - tool names are type-checked for autocomplete. If a tool takes longer than its timeout, it aborts and returns a tool-error so the model can respond or retry. Can be used alongside abortSignal.',
+        'Timeout in milliseconds. Can be specified as a number or as an object with totalMs, stepMs, firstChunkMs, chunkMs, toolMs, and/or tools properties. totalMs sets the total timeout for the entire call. stepMs sets the timeout for each individual step (LLM call). firstChunkMs sets the timeout until the first content-bearing output in each step; metadata, stream starts, empty deltas, raw chunks, and transport activity do not satisfy it. chunkMs sets the timeout between content-bearing output chunks after output has started; non-content chunks do not reset it. toolMs sets the default timeout for all tool executions. tools sets per-tool timeout overrides using the pattern {toolName}Ms (e.g. weatherMs, slowApiMs) that take precedence over toolMs - tool names are type-checked for autocomplete. If a tool takes longer than its timeout, it aborts and returns a tool-error so the model can respond or retry. Can be used alongside abortSignal.',
     },
     {
       name: 'headers',
@@ -4020,7 +4020,7 @@ To see `streamText` in action, check out [these examples](#examples).
     },
     {
       name: 'pipeUIMessageStreamToResponse',
-      type: '(response: ServerResponse, options?: ResponseInit & UIMessageStreamOptions) => void',
+      type: '(response: ServerResponse, options?: ResponseInit & UIMessageStreamOptions) => Promise<void>',
       description:
         'Writes UI message stream output to a Node.js response-like object.',
       properties: [
@@ -4051,7 +4051,7 @@ To see `streamText` in action, check out [these examples](#examples).
     },
     {
       name: 'pipeTextStreamToResponse',
-      type: '(response: ServerResponse, init?: ResponseInit) => void',
+      type: '(response: ServerResponse, init?: ResponseInit) => Promise<void>',
       description:
         'Writes text delta output to a Node.js response-like object. It sets a `Content-Type` header to `text/plain; charset=utf-8` and writes each text delta as a separate chunk.',
       properties: [

package/docs/07-reference/01-ai-sdk-core/16-tool-loop-agent.mdx

@@ -246,8 +246,9 @@ To see `ToolLoopAgent` in action, check out [these examples](#examples).
             },
             {
               name: 'timeout',
-              type: 'number | { totalMs?: number; stepMs?: number; chunkMs?: number } | undefined',
-              description: 'Timeout configuration for the generation.',
+              type: 'number | { totalMs?: number; stepMs?: number; firstChunkMs?: number; chunkMs?: number } | undefined',
+              description:
+                'Timeout configuration for the generation. firstChunkMs and chunkMs are only enforced by stream().',
             },
             {
               name: 'headers',
@@ -360,8 +361,9 @@ To see `ToolLoopAgent` in action, check out [these examples](#examples).
             },
             {
               name: 'timeout',
-              type: 'number | { totalMs?: number; stepMs?: number; chunkMs?: number } | undefined',
-              description: 'Timeout configuration for the generation.',
+              type: 'number | { totalMs?: number; stepMs?: number; firstChunkMs?: number; chunkMs?: number } | undefined',
+              description:
+                'Timeout configuration for the generation. firstChunkMs and chunkMs are only enforced by stream().',
             },
             {
               name: 'headers',
@@ -712,10 +714,10 @@ const result = await agent.generate({
     },
     {
       name: 'timeout',
-      type: 'number | { totalMs?: number; stepMs?: number; chunkMs?: number }',
+      type: 'number | { totalMs?: number; stepMs?: number; firstChunkMs?: number; chunkMs?: number }',
       isOptional: true,
       description:
-        'Timeout in milliseconds. Can be specified as a number or as an object with totalMs, stepMs, and/or chunkMs properties. The call will be aborted if it takes longer than the specified timeout. Can be used alongside abortSignal.',
+        'Timeout in milliseconds. Can be specified as a number or as an object with totalMs, stepMs, firstChunkMs, and/or chunkMs properties. firstChunkMs and chunkMs have no effect on generate(); they are streaming-only and are enforced by stream(). Can be used alongside abortSignal.',
     },
     {
       name: 'experimental_sandbox',
@@ -828,10 +830,10 @@ for await (const chunk of stream.textStream) {
     },
     {
       name: 'timeout',
-      type: 'number | { totalMs?: number; stepMs?: number; chunkMs?: number }',
+      type: 'number | { totalMs?: number; stepMs?: number; firstChunkMs?: number; chunkMs?: number }',
       isOptional: true,
       description:
-        'Timeout in milliseconds. Can be specified as a number or as an object with totalMs, stepMs, and/or chunkMs properties. The call will be aborted if it takes longer than the specified timeout. Can be used alongside abortSignal.',
+        'Timeout in milliseconds. Can be specified as a number or as an object with totalMs, stepMs, firstChunkMs, and/or chunkMs properties. firstChunkMs limits the wait for the first content-bearing output in each model-call step. chunkMs limits gaps between later content-bearing output chunks. Can be used alongside abortSignal.',
     },
     {
       name: 'experimental_sandbox',

package/docs/07-reference/02-ai-sdk-ui/42-pipe-ui-message-stream-to-response.mdx

@@ -17,7 +17,7 @@ The `pipeUIMessageStreamToResponse` function pipes streaming data to a Node.js S
 ## Example
 
 ```tsx
-pipeUIMessageStreamToResponse({
+await pipeUIMessageStreamToResponse({
   response: serverResponse,
   status: 200,
   statusText: 'OK',
@@ -75,3 +75,8 @@ pipeUIMessageStreamToResponse({
     },
   ]}
 />
+
+### Returns
+
+A `Promise<void>` that resolves when the stream has been written to the response
+and rejects when reading or writing the stream fails.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ai",
-  "version": "7.0.34",
+  "version": "7.0.35",
   "type": "module",
   "description": "AI SDK by Vercel - build apps like ChatGPT, Claude, Gemini, and more with a single interface for any model using the Vercel AI Gateway or go direct to OpenAI, Anthropic, Google, or any other model provider.",
   "license": "Apache-2.0",
@@ -42,7 +42,7 @@
     }
   },
   "dependencies": {
-    "@ai-sdk/gateway": "4.0.26",
+    "@ai-sdk/gateway": "4.0.27",
     "@ai-sdk/provider": "4.0.3",
     "@ai-sdk/provider-utils": "5.0.12"
   },

package/src/agent/agent.ts

@@ -70,7 +70,10 @@ export type AgentCallParameters<
     abortSignal?: AbortSignal;
 
     /**
-     * Timeout in milliseconds. Can be specified as a number or as an object with `totalMs`.
+     * Timeout in milliseconds. Can be specified as a number or as an object
+     * with total, per-step, first-content, inter-content, and tool timeouts.
+     * First-content and inter-content timeouts are only enforced by streaming
+     * calls.
      */
     timeout?: TimeoutConfiguration<TOOLS>;
 

package/src/agent/pipe-agent-ui-stream-to-response.ts

@@ -66,7 +66,7 @@ export async function pipeAgentUIStreamToResponse<
   UIMessageStreamOptions<
     UIMessage<MESSAGE_METADATA, never, InferUITools<TOOLS>>
   >): Promise<void> {
-  pipeUIMessageStreamToResponse({
+  return pipeUIMessageStreamToResponse({
     response,
     headers,
     status,

package/src/generate-object/stream-object-result.ts

@@ -85,7 +85,10 @@ export interface StreamObjectResult<PARTIAL, RESULT, ELEMENT_STREAM> {
    * @param response A Node.js response-like object (ServerResponse).
    * @param init Optional headers, status code, and status text.
    */
-  pipeTextStreamToResponse(response: ServerResponse, init?: ResponseInit): void;
+  pipeTextStreamToResponse(
+    response: ServerResponse,
+    init?: ResponseInit,
+  ): Promise<void>;
 
   /**
    * Creates a simple text stream response.

package/src/generate-object/stream-object.ts

@@ -964,7 +964,7 @@ class DefaultStreamObjectResult<
   }
 
   pipeTextStreamToResponse(response: ServerResponse, init?: ResponseInit) {
-    pipeTextStreamToResponse({
+    return pipeTextStreamToResponse({
       response,
       stream: this.textStream,
       ...init,

package/src/generate-text/generate-text-events.ts

@@ -56,7 +56,8 @@ export type GenerateTextStartEvent<
 
   /**
    * Timeout configuration for the generation.
-   * Can be a number (milliseconds) or an object with totalMs, stepMs, chunkMs, toolMs, and per-tool overrides via tools.
+   * Can be a number (milliseconds) or an object with totalMs, stepMs,
+   * firstChunkMs (streaming only), chunkMs, toolMs, and per-tool overrides via tools.
    */
   readonly timeout: TimeoutConfiguration<TOOLS> | undefined;
 

package/src/generate-text/stream-text-result.ts

@@ -381,7 +381,7 @@ export interface StreamTextResult<
   pipeUIMessageStreamToResponse<UI_MESSAGE extends UIMessage>(
     response: ServerResponse,
     options?: UIMessageStreamResponseInit & UIMessageStreamOptions<UI_MESSAGE>,
-  ): void;
+  ): Promise<void>;
 
   /**
    * Writes text delta output to a Node.js response-like object.
@@ -395,7 +395,10 @@ export interface StreamTextResult<
    *   `pipeTextStreamToResponse` helpers from `'ai'` with `result.stream`
    *   instead. This method will be removed in the next major release.
    */
-  pipeTextStreamToResponse(response: ServerResponse, init?: ResponseInit): void;
+  pipeTextStreamToResponse(
+    response: ServerResponse,
+    init?: ResponseInit,
+  ): Promise<void>;
 
   /**
    * Converts the result to a streamed response object with a stream data part stream.

package/src/generate-text/stream-text.ts

@@ -34,6 +34,7 @@ import { prepareTools } from '../prompt/prepare-tools';
 import type { Prompt } from '../prompt/prompt';
 import {
   getChunkTimeoutMs,
+  getFirstChunkTimeoutMs,
   getStepTimeoutMs,
   getTotalTimeoutMs,
   type RequestOptions,
@@ -157,28 +158,29 @@ const originalGenerateCallId = createIdGenerator({
   size: 24,
 });
 
-// chunk types that count as model output; used to distinguish empty
-// incomplete streams from incomplete streams with partial results.
-// exhaustive so that new chunk types must be classified explicitly:
+// Chunk types that contain semantic model output. This classification is used
+// for first-content and inter-content timeouts as well as to distinguish empty
+// incomplete streams from incomplete streams with partial results. It is
+// exhaustive so that new chunk types must be classified explicitly.
 const isOutputChunkType = {
   file: true,
-  custom: true,
-  source: true,
-  'text-start': true,
-  'text-end': true,
+  custom: false,
+  source: false,
+  'text-start': false,
+  'text-end': false,
   'text-delta': true,
-  'reasoning-start': true,
-  'reasoning-end': true,
+  'reasoning-start': false,
+  'reasoning-end': false,
   'reasoning-delta': true,
   'reasoning-file': true,
-  'tool-input-start': true,
-  'tool-input-end': true,
+  'tool-input-start': false,
+  'tool-input-end': false,
   'tool-input-delta': true,
-  'tool-approval-request': true,
-  'tool-approval-response': true,
+  'tool-approval-request': false,
+  'tool-approval-response': false,
   'tool-call': true,
-  'tool-result': true,
-  'tool-error': true,
+  'tool-result': false,
+  'tool-error': false,
   'tool-execution-end': false,
   'model-call-start': false,
   'model-call-response-metadata': false,
@@ -187,6 +189,26 @@ const isOutputChunkType = {
   raw: false,
 } as const satisfies Record<ExecuteToolsStreamPart['type'], boolean>;
 
+function isOutputChunk(chunk: ExecuteToolsStreamPart): boolean {
+  if (!isOutputChunkType[chunk.type]) {
+    return false;
+  }
+
+  switch (chunk.type) {
+    case 'text-delta':
+    case 'reasoning-delta':
+      return chunk.text.length > 0;
+    case 'tool-input-delta':
+      return chunk.delta.length > 0;
+    case 'file':
+    case 'reasoning-file':
+    case 'tool-call':
+      return true;
+    default:
+      return false;
+  }
+}
+
 export type StreamTextInclude = {
   /**
    * Whether to retain the request body in step results.
@@ -716,9 +738,12 @@ export function streamText<
   }): StreamTextResult<TOOLS, RUNTIME_CONTEXT, OUTPUT> {
   const totalTimeoutMs = getTotalTimeoutMs(timeout);
   const stepTimeoutMs = getStepTimeoutMs(timeout);
+  const firstChunkTimeoutMs = getFirstChunkTimeoutMs(timeout);
   const chunkTimeoutMs = getChunkTimeoutMs(timeout);
   const stepAbortController =
     stepTimeoutMs != null ? new AbortController() : undefined;
+  const firstChunkAbortController =
+    firstChunkTimeoutMs != null ? new AbortController() : undefined;
   const chunkAbortController =
     chunkTimeoutMs != null ? new AbortController() : undefined;
   const resolvedOnStart = onStart ?? experimental_onStart;
@@ -742,10 +767,13 @@ export function streamText<
       abortSignal,
       totalTimeoutMs,
       stepAbortController?.signal,
+      firstChunkAbortController?.signal,
       chunkAbortController?.signal,
     ),
     stepTimeoutMs,
     stepAbortController,
+    firstChunkTimeoutMs,
+    firstChunkAbortController,
     chunkTimeoutMs,
     chunkAbortController,
     instructions,
@@ -930,6 +958,10 @@ class DefaultStreamTextResult<
 
   private readonly addStream: (
     stream: ReadableStream<TextStreamPart<TOOLS>>,
+    callbacks?: {
+      onError?: (error: unknown) => void;
+      onCancel?: () => void;
+    },
   ) => void;
 
   private readonly closeStream: () => void;
@@ -951,6 +983,8 @@ class DefaultStreamTextResult<
     abortSignal,
     stepTimeoutMs,
     stepAbortController,
+    firstChunkTimeoutMs,
+    firstChunkAbortController,
     chunkTimeoutMs,
     chunkAbortController,
     instructions,
@@ -1000,6 +1034,8 @@ class DefaultStreamTextResult<
     abortSignal: AbortSignal | undefined;
     stepTimeoutMs: number | undefined;
     stepAbortController: AbortController | undefined;
+    firstChunkTimeoutMs: number | undefined;
+    firstChunkAbortController: AbortController | undefined;
     chunkTimeoutMs: number | undefined;
     chunkAbortController: AbortController | undefined;
     toolsContext: InferToolSetContext<TOOLS>;
@@ -1771,7 +1807,32 @@ class DefaultStreamTextResult<
           timeoutMs: stepTimeoutMs,
         });
 
-        // Set up chunk timeout tracking (will be reset on each chunk)
+        // The first-content timeout is armed when the provider response stream
+        // starts and is cleared by the first semantic output chunk.
+        let firstChunkTimeoutId: ReturnType<typeof setTimeout> | undefined =
+          undefined;
+
+        function startFirstChunkTimeout() {
+          if (abortSignal?.aborted) {
+            return;
+          }
+
+          firstChunkTimeoutId = setAbortTimeout({
+            abortController: firstChunkAbortController,
+            label: 'First chunk',
+            timeoutMs: firstChunkTimeoutMs,
+          });
+        }
+
+        function clearFirstChunkTimeout() {
+          if (firstChunkTimeoutId != null) {
+            clearTimeout(firstChunkTimeoutId);
+            firstChunkTimeoutId = undefined;
+          }
+        }
+
+        // Chunk timeout tracking starts after semantic output begins and is
+        // reset only by subsequent semantic output chunks.
         let chunkTimeoutId: ReturnType<typeof setTimeout> | undefined =
           undefined;
 
@@ -1799,12 +1860,24 @@ class DefaultStreamTextResult<
           }
         }
 
+        function clearStepTimeouts() {
+          clearStepTimeout();
+          clearFirstChunkTimeout();
+          clearChunkTimeout();
+        }
+
+        function cleanupStepTimeouts() {
+          abortSignal?.removeEventListener('abort', cleanupStepTimeouts);
+          clearStepTimeouts();
+        }
+
         // The step's stream is registered lazily and consumed long after this
-        // function returns, so the step timer must stay armed past setup. When
-        // the merged abort signal fires (any step/chunk/total timeout or caller
-        // abort), drop both step-scoped timers so neither outlives the step.
-        abortSignal?.addEventListener('abort', clearStepTimeout);
-        abortSignal?.addEventListener('abort', clearChunkTimeout);
+        // function returns, so its timers must stay armed past setup. When the
+        // merged abort signal fires, drop all step-scoped timers so none
+        // outlives the step.
+        abortSignal?.addEventListener('abort', cleanupStepTimeouts, {
+          once: true,
+        });
 
         try {
           stepFinish = new DelayedPromise<void>();
@@ -1967,6 +2040,8 @@ class DefaultStreamTextResult<
             ),
           );
 
+          startFirstChunkTimeout();
+
           const streamAfterToolCallbackInvocation =
             invokeToolCallbacksFromStream({
               stream: languageModelStream,
@@ -2076,8 +2151,6 @@ class DefaultStreamTextResult<
                 TextStreamPart<TOOLS>
               >({
                 async transform(chunk, controller): Promise<void> {
-                  resetChunkTimeout();
-
                   if (chunk.type === 'model-call-start') {
                     warnings = chunk.warnings;
                     return; // stream start chunks are sent immediately and do not count as first chunk
@@ -2096,8 +2169,14 @@ class DefaultStreamTextResult<
 
                   const chunkType = chunk.type;
 
-                  if (isOutputChunkType[chunkType]) {
+                  if (isOutputChunk(chunk)) {
+                    if (!hasReceivedOutputChunk) {
+                      // Clear before forwarding the first output so a timeout
+                      // cannot race with already-visible generated content.
+                      clearFirstChunkTimeout();
+                    }
                     hasReceivedOutputChunk = true;
+                    resetChunkTimeout();
                   }
 
                   switch (chunkType) {
@@ -2217,8 +2296,7 @@ class DefaultStreamTextResult<
                       }),
                     });
 
-                    clearStepTimeout();
-                    clearChunkTimeout();
+                    cleanupStepTimeouts();
                     self.closeStream();
                     return;
                   }
@@ -2298,9 +2376,8 @@ class DefaultStreamTextResult<
                     }
                   }
 
-                  // Clear the step and chunk timeouts before the next step is started
-                  clearStepTimeout();
-                  clearChunkTimeout();
+                  // Clear this step's timeouts before the next step is started.
+                  cleanupStepTimeouts();
 
                   if (
                     // Continue if:
@@ -2345,12 +2422,15 @@ class DefaultStreamTextResult<
                 },
               }),
             ),
+            {
+              onError: cleanupStepTimeouts,
+              onCancel: cleanupStepTimeouts,
+            },
           );
         } catch (error) {
           // Setup failed before the stream was registered, so neither the
           // stream's flush nor an abort will clear the timers — clear them here.
-          clearStepTimeout();
-          clearChunkTimeout();
+          cleanupStepTimeouts();
           throw error;
         }
       }
@@ -2678,7 +2758,7 @@ class DefaultStreamTextResult<
       ...init
     }: UIMessageStreamResponseInit & UIMessageStreamOptions<UI_MESSAGE> = {},
   ) {
-    pipeUIMessageStreamToResponse({
+    return pipeUIMessageStreamToResponse({
       response,
       stream: this.toUIMessageStream({
         originalMessages,
@@ -2696,7 +2776,7 @@ class DefaultStreamTextResult<
   }
 
   pipeTextStreamToResponse(response: ServerResponse, init?: ResponseInit) {
-    pipeTextStreamToResponse({
+    return pipeTextStreamToResponse({
       response,
       stream: this.textStream,
       ...init,

package/src/prompt/index.ts

@@ -10,6 +10,7 @@ export type CallSettings = LanguageModelCallOptions &
 export {
   getTotalTimeoutMs,
   getStepTimeoutMs,
+  getFirstChunkTimeoutMs,
   getChunkTimeoutMs,
   getToolTimeoutMs,
 } from './request-options';