@ai-sdk/anthropic 4.0.46 → 4.0.48

package/docs/05-anthropic.mdx

@@ -233,9 +233,14 @@ const result = streamText({
 });
 ```
 
+For `claude-fable-5-1`, the default `"auto"` mode uses native structured
+outputs through `output_config.format`. Fable 5.1 rejects forced tool use, so
+do not use `"jsonTool"` or a required or named tool choice to implement
+structured output with this model.
+
 ### Effort
 
-Anthropic introduced an `effort` option with `claude-opus-4-5` that affects thinking, text responses, and function calls. Effort defaults to `high` and you can set it to `medium` or `low` to save tokens and to lower time-to-last-token latency (TTLT). `claude-opus-4-7`, `claude-opus-4-8`, `claude-opus-5`, `claude-fable-5`, and `claude-sonnet-5` additionally support `xhigh` for maximum reasoning effort.
+Anthropic introduced an `effort` option with `claude-opus-4-5` that affects thinking, text responses, and function calls. Effort defaults to `high` and you can set it to `medium` or `low` to save tokens and to lower time-to-last-token latency (TTLT). `claude-opus-4-7`, `claude-opus-4-8`, `claude-opus-5`, `claude-fable-5`, `claude-fable-5-1`, and `claude-sonnet-5` additionally support `xhigh` for maximum reasoning effort.
 
 On `claude-opus-5`, thinking can only be disabled at effort levels up to and including `high`. When you combine `thinking: { type: 'disabled' }` with `effort: 'xhigh'` or `effort: 'max'`, the AI SDK lowers the effort to `high` and emits a warning instead of sending a request that the API would reject.
 
@@ -484,6 +489,56 @@ console.log(text);
   `display: "summarized"` to restore visible progress during thinking.
 </Note>
 
+##### Thinking Updates
+
+Use `display: 'updates'` with `claude-fable-5-1` to stream thinking summaries
+between tool calls. The provider adds the required
+`thinking-display-updates-2026-08-18` beta header automatically:
+
+```ts highlight="12-16"
+import { anthropic, AnthropicLanguageModelOptions } from '@ai-sdk/anthropic';
+import { streamText, tool } from 'ai';
+import { z } from 'zod';
+
+const result = streamText({
+  model: anthropic('claude-fable-5-1'),
+  tools: {
+    weather: tool({
+      inputSchema: z.object({ city: z.string() }),
+    }),
+  },
+  providerOptions: {
+    anthropic: {
+      thinking: { type: 'adaptive', display: 'updates' },
+    } satisfies AnthropicLanguageModelOptions,
+  },
+  prompt: 'Compare the weather in San Francisco and New York.',
+});
+```
+
+##### Thinking Binding Controls
+
+Fable 5.1 can recover from a thinking-block prefix mismatch by dropping the
+mismatched block. Set `blockBinding.prefixMismatchBehavior` to `drop_block`.
+You can provide block binding by itself to preserve the model's default
+thinking mode, or combine it with adaptive thinking.
+
+```ts highlight="7-11"
+const { text } = await generateText({
+  model: anthropic('claude-fable-5-1'),
+  prompt: 'Continue from this conversation.',
+  providerOptions: {
+    anthropic: {
+      thinking: {
+        blockBinding: {
+          prefixMismatchBehavior: 'drop_block',
+        },
+      },
+    } satisfies AnthropicLanguageModelOptions,
+  },
+});
+```
+
 #### Budget-Based Thinking
 
 For earlier models (`claude-opus-4-20250514`, `claude-sonnet-4-20250514`, `claude-sonnet-4-5-20250929`),
@@ -1279,6 +1334,59 @@ const result = await generateText({
 
 This sends `tool_reference` blocks to Anthropic, which loads the corresponding deferred tool schemas into Claude's context.
 
+### Mid-Conversation System Controls
+
+With `claude-fable-5-1`, a mid-conversation system message can be cleared
+before the next user message with `clearAt: 'next_user_message'`, or set the
+effort for the next turn. The provider adds the required
+`mid-conversation-system-clear-at-2026-08-21` and
+`mid-conversation-effort-2026-08-01` beta headers automatically.
+
+```ts highlight="17-25"
+import { anthropic } from '@ai-sdk/anthropic';
+import { generateText } from 'ai';
+
+const result = await generateText({
+  model: anthropic('claude-fable-5-1'),
+  allowSystemInMessages: true,
+  messages: [
+    {
+      role: 'user',
+      content: 'Draft a migration plan.',
+    },
+    {
+      role: 'assistant',
+      content: 'First, inventory the public API.',
+    },
+    {
+      role: 'system',
+      content: 'For the next turn, verify every compatibility claim.',
+      providerOptions: {
+        anthropic: {
+          clearAt: 'next_user_message',
+          effort: 'high',
+        },
+      },
+    },
+    {
+      role: 'user',
+      content: 'Complete the plan.',
+    },
+  ],
+});
+```
+
+An effort-only system message can use an empty content array at the API level;
+in an AI SDK prompt, set `content: ''`. The provider serializes it as
+`content: []`.
+
+<Note>
+  `clearAt` and per-message `effort` apply only to mid-conversation system
+  messages. When used on the initial system message, the provider ignores them
+  and emits a warning. Use the top-level `effort` provider option to configure
+  effort for the full request.
+</Note>
+
 ### Mid-Conversation Tool Changes
 
 With `claude-opus-4-8`, you can add or remove tools between turns of a conversation without invalidating the prompt cache. Attach `toolChanges` to a system message that appears mid-conversation (right before an assistant message or at the end of the messages). The required `mid-conversation-tool-changes-2026-07-01` beta header is added automatically.
@@ -1760,6 +1868,7 @@ and the `mediaType` should be set to `'application/pdf'`.
 | ------------------- | ----------- | ----------------- | ---------- | ------------ | ---------- | ----------- | ---------- |
 | `claude-opus-5`     | <Check />   | <Check />         | <Check />  | <Check />    | <Check />  | <Check />   | <Check />  |
 | `claude-sonnet-5`   | <Check />   | <Check />         | <Check />  | <Check />    | <Check />  | <Check />   | <Check />  |
+| `claude-fable-5-1`  | <Check />   | <Check />         | <Check />  | <Check />    | <Check />  | <Check />   | <Check />  |
 | `claude-fable-5`    | <Check />   | <Check />         | <Check />  | <Check />    | <Check />  | <Check />   | <Check />  |
 | `claude-opus-4-8`   | <Check />   | <Check />         | <Check />  | <Check />    | <Check />  | <Check />   | <Check />  |
 | `claude-opus-4-7`   | <Check />   | <Check />         | <Check />  | <Check />    | <Check />  | <Check />   | <Check />  |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/anthropic",
-  "version": "4.0.46",
+  "version": "4.0.48",
   "type": "module",
   "license": "Apache-2.0",
   "sideEffects": false,
@@ -35,8 +35,8 @@
     }
   },
   "dependencies": {
-    "@ai-sdk/provider": "4.0.9",
-    "@ai-sdk/provider-utils": "5.0.34"
+    "@ai-sdk/provider": "4.0.10",
+    "@ai-sdk/provider-utils": "5.0.36"
   },
   "devDependencies": {
     "@ai-sdk/test-server": "2.0.1",

package/src/anthropic-api.ts

@@ -24,6 +24,10 @@ export type AnthropicCacheControl = {
 export interface AnthropicSystemMessage {
   role: 'system';
   content: Array<AnthropicTextContent | AnthropicToolChangeContent>;
+  clear_at?: 'next_user_message';
+  output_config?: {
+    effort: 'low' | 'medium' | 'high' | 'xhigh' | 'max';
+  };
 }
 
 /**

package/src/anthropic-files.ts

@@ -49,6 +49,8 @@ export class AnthropicFiles implements FilesV4 {
     data,
     mediaType,
     filename,
+    abortSignal,
+    headers,
   }: FilesV4UploadFileCallOptions): Promise<FilesV4UploadFileResult> {
     const fileBytes = convertInlineFileDataToUint8Array(data);
 
@@ -63,14 +65,17 @@ export class AnthropicFiles implements FilesV4 {
 
     const { value: response } = await postFormDataToApi({
       url: `${this.config.baseURL}/files`,
-      headers: combineHeaders(this.config.headers(), {
-        'anthropic-beta': 'files-api-2025-04-14',
-      }),
+      headers: combineHeaders(
+        this.config.headers(),
+        { 'anthropic-beta': 'files-api-2025-04-14' },
+        headers,
+      ),
       formData,
       failedResponseHandler: anthropicFailedResponseHandler,
       successfulResponseHandler: createJsonResponseHandler(
         anthropicUploadFileResponseSchema,
       ),
+      abortSignal,
       fetch: this.config.fetch,
     });
 

package/src/anthropic-language-model-options.ts

@@ -21,6 +21,7 @@ export type AnthropicModelId =
   | 'claude-opus-4-8'
   | 'claude-opus-5'
   | 'claude-fable-5'
+  | 'claude-fable-5-1'
   | 'claude-sonnet-5'
   | (string & {});
 
@@ -70,6 +71,22 @@ export type AnthropicFilePartProviderOptions = z.infer<
  * Anthropic provider options for system messages.
  */
 export const anthropicSystemMessageProviderOptions = z.object({
+  /**
+   * Controls when an ephemeral mid-conversation system message is cleared.
+   *
+   * Only supported on system messages that appear mid-conversation.
+   * The required `mid-conversation-system-clear-at-2026-08-21` beta is
+   * added automatically.
+   */
+  clearAt: z.literal('next_user_message').optional(),
+
+  /**
+   * Sets the model effort for the turn that follows this mid-conversation
+   * system message. The required `mid-conversation-effort-2026-08-01` beta
+   * is added automatically.
+   */
+  effort: z.enum(['low', 'medium', 'high', 'xhigh', 'max']).optional(),
+
   /**
    * Mid-conversation tool changes. Adds or removes tools from the
    * conversation's tool set between turns without invalidating the prompt
@@ -128,24 +145,42 @@ export const anthropicLanguageModelOptions = z.object({
    * Requires a minimum budget of 1,024 tokens and counts towards the `max_tokens` limit.
    */
   thinking: z
-    .discriminatedUnion('type', [
-      z.object({
-        /** for Sonnet 4.6, Opus 4.6, and newer models */
-        type: z.literal('adaptive'),
-        /**
-         * Controls whether thinking content is included in the response.
-         * - `"omitted"`: Thinking blocks are present but text is empty (default for Opus 4.7+).
-         * - `"summarized"`: Thinking content is returned. Required to see reasoning output.
-         */
-        display: z.enum(['omitted', 'summarized']).optional(),
-      }),
-      z.object({
-        /** for models before Opus 4.6, except Sonnet 4.6 still supports it */
-        type: z.literal('enabled'),
-        budgetTokens: z.number().optional(),
-      }),
+    .union([
+      z.discriminatedUnion('type', [
+        z.object({
+          /** for Sonnet 4.6, Opus 4.6, and newer models */
+          type: z.literal('adaptive'),
+          /**
+           * Controls whether thinking content is included in the response.
+           * - `"omitted"`: Thinking blocks are present but text is empty (default for Opus 4.7+).
+           * - `"summarized"`: Thinking content is returned.
+           * - `"updates"`: Thinking updates are returned between tool calls.
+           */
+          display: z.enum(['omitted', 'summarized', 'updates']).optional(),
+          /**
+           * Controls how thinking blocks are bound to an existing thinking
+           * prefix. Requires the `thinking-binding-controls-2026-08-01` beta.
+           */
+          blockBinding: z
+            .object({
+              prefixMismatchBehavior: z.literal('drop_block'),
+            })
+            .optional(),
+        }),
+        z.object({
+          /** for models before Opus 4.6, except Sonnet 4.6 still supports it */
+          type: z.literal('enabled'),
+          budgetTokens: z.number().optional(),
+        }),
+        z.object({
+          type: z.literal('disabled'),
+        }),
+      ]),
       z.object({
-        type: z.literal('disabled'),
+        type: z.never().optional(),
+        blockBinding: z.object({
+          prefixMismatchBehavior: z.literal('drop_block'),
+        }),
       }),
     ])
     .optional(),

package/src/anthropic-language-model.ts

@@ -535,10 +535,17 @@ export class AnthropicLanguageModel implements LanguageModelV4 {
     const thinkingType = anthropicOptions?.thinking?.type;
     const isThinking =
       thinkingType === 'enabled' || thinkingType === 'adaptive';
+    const thinkingBlockBinding =
+      anthropicOptions?.thinking != null &&
+      'blockBinding' in anthropicOptions.thinking
+        ? anthropicOptions.thinking.blockBinding
+        : undefined;
     // `disabled` must still be forwarded to the API: some models (e.g. Sonnet 5)
     // default thinking on, so omitting it would leave thinking enabled and
-    // consume the max_tokens budget.
-    const sendThinking = isThinking || thinkingType === 'disabled';
+    // consume the max_tokens budget. Binding-only recovery requests must also
+    // send a thinking object without a type.
+    const sendThinking =
+      isThinking || thinkingType === 'disabled' || thinkingBlockBinding != null;
     let thinkingBudget =
       thinkingType === 'enabled'
         ? anthropicOptions?.thinking?.budgetTokens
@@ -564,9 +571,15 @@ export class AnthropicLanguageModel implements LanguageModelV4 {
       // provider specific settings:
       ...(sendThinking && {
         thinking: {
-          type: thinkingType,
+          ...(thinkingType != null && { type: thinkingType }),
           ...(thinkingBudget != null && { budget_tokens: thinkingBudget }),
           ...(thinkingDisplay != null && { display: thinkingDisplay }),
+          ...(thinkingBlockBinding != null && {
+            block_binding: {
+              prefix_mismatch_behavior:
+                thinkingBlockBinding.prefixMismatchBehavior,
+            },
+          }),
         },
       }),
       ...((anthropicOptions?.effort ||
@@ -843,6 +856,14 @@ export class AnthropicLanguageModel implements LanguageModelV4 {
       betas.add('fast-mode-2026-02-01');
     }
 
+    if (thinkingDisplay === 'updates') {
+      betas.add('thinking-display-updates-2026-08-18');
+    }
+
+    if (thinkingBlockBinding != null) {
+      betas.add('thinking-binding-controls-2026-08-01');
+    }
+
     if (anthropicOptions?.fallbacks === 'default') {
       betas.add('server-side-fallback-2026-07-01');
     } else if (

package/src/convert-to-anthropic-prompt.ts

@@ -144,8 +144,12 @@ export async function convertToAnthropicPrompt({
 
     switch (type) {
       case 'system': {
-        const content: AnthropicSystemMessage['content'] = [];
-        let toolChangeCount = 0;
+        const convertedMessages: Array<{
+          content: AnthropicSystemMessage['content'];
+          clearAt?: 'next_user_message';
+          effort?: 'low' | 'medium' | 'high' | 'xhigh' | 'max';
+          toolChangeCount: number;
+        }> = [];
 
         for (const { content: text, providerOptions } of block.messages) {
           const systemMessageOptions = await parseProviderOptions({
@@ -154,10 +158,16 @@ export async function convertToAnthropicPrompt({
             schema: anthropicSystemMessageProviderOptions,
           });
           const toolChanges = systemMessageOptions?.toolChanges ?? [];
-
-          // A system message that only carries tool changes may have empty
-          // text; do not emit an empty text block for it.
-          if (text !== '' || toolChanges.length === 0) {
+          const content: AnthropicSystemMessage['content'] = [];
+
+          // A system message that only carries message-level controls may have
+          // empty text; do not emit an empty text block for it.
+          if (
+            text !== '' ||
+            (toolChanges.length === 0 &&
+              systemMessageOptions?.clearAt == null &&
+              systemMessageOptions?.effort == null)
+          ) {
             content.push({
               type: 'text' as const,
               text,
@@ -169,7 +179,6 @@ export async function convertToAnthropicPrompt({
           }
 
           for (const toolChange of toolChanges) {
-            toolChangeCount++;
             content.push({
               type: toolChange.type,
               tool: {
@@ -178,6 +187,13 @@ export async function convertToAnthropicPrompt({
               },
             } satisfies AnthropicToolChangeContent);
           }
+
+          convertedMessages.push({
+            content,
+            clearAt: systemMessageOptions?.clearAt,
+            effort: systemMessageOptions?.effort,
+            toolChangeCount: toolChanges.length,
+          });
         }
 
         // The first block becomes the top-level system prompt. Later system
@@ -185,7 +201,17 @@ export async function convertToAnthropicPrompt({
         // tool changes (which are only valid mid-conversation), and otherwise
         // only when a top-level system prompt already exists (preserving the
         // existing hoisting behavior for plain text).
-        if (i === 0 || (system == null && toolChangeCount === 0)) {
+        const toolChangeCount = convertedMessages.reduce(
+          (count, message) => count + message.toolChangeCount,
+          0,
+        );
+        const hasInlineSystemOptions = convertedMessages.some(
+          message => message.clearAt != null || message.effort != null,
+        );
+        if (
+          i === 0 ||
+          (system == null && toolChangeCount === 0 && !hasInlineSystemOptions)
+        ) {
           if (toolChangeCount > 0) {
             warnings.push({
               type: 'other',
@@ -195,14 +221,45 @@ export async function convertToAnthropicPrompt({
                 'The tool changes have been ignored.',
             });
           }
-          system = content.filter(
-            (part): part is AnthropicTextContent => part.type === 'text',
+
+          for (const message of convertedMessages) {
+            if (message.clearAt != null || message.effort != null) {
+              warnings.push({
+                type: 'other',
+                message:
+                  'clearAt and effort on the initial system message are not supported by Anthropic. ' +
+                  'These options have been ignored.',
+              });
+            }
+          }
+
+          system = convertedMessages.flatMap(message =>
+            message.content.filter(
+              (part): part is AnthropicTextContent => part.type === 'text',
+            ),
           );
         } else {
-          messages.push({ role: 'system', content });
           betas.add('mid-conversation-system-2026-04-07');
-          if (toolChangeCount > 0) {
-            betas.add('mid-conversation-tool-changes-2026-07-01');
+
+          for (const message of convertedMessages) {
+            messages.push({
+              role: 'system',
+              content: message.content,
+              ...(message.clearAt != null && { clear_at: message.clearAt }),
+              ...(message.effort != null && {
+                output_config: { effort: message.effort },
+              }),
+            });
+
+            if (message.toolChangeCount > 0) {
+              betas.add('mid-conversation-tool-changes-2026-07-01');
+            }
+            if (message.clearAt != null) {
+              betas.add('mid-conversation-system-clear-at-2026-08-21');
+            }
+            if (message.effort != null) {
+              betas.add('mid-conversation-effort-2026-08-01');
+            }
           }
         }