@ai-sdk/anthropic 4.0.55 → 4.0.57

package/docs/05-anthropic.mdx

@@ -1893,3 +1893,52 @@ and the `mediaType` should be set to `'application/pdf'`.
   of available models. The table above lists popular models. You can also pass
   any available provider model ID as a string if needed.
 </Note>
+
+## Evaluation Models
+
+Create an experimental evaluation model with `anthropic.evaluationModel(modelId)`.
+It uses the Messages API's structured-output support for Choice, Score, and Boolean
+questions. `claude-haiku-4-5-20251001` is one supported model.
+
+```ts
+import { anthropic } from '@ai-sdk/anthropic';
+import { experimental_evaluate } from 'ai';
+
+const { answers } = await experimental_evaluate({
+  model: anthropic.evaluationModel('claude-haiku-4-5-20251001'),
+  state: 'I was charged twice.',
+  questions: {
+    requestsRefund: {
+      type: 'boolean',
+      instructions: 'Is the customer requesting money back?',
+    },
+    department: {
+      type: 'choice',
+      instructions: 'Which team should handle this?',
+      criteria: { billing: 'Charges and refunds', support: 'Other requests' },
+    },
+  },
+});
+```
+
+Choice labels are preserved exactly, and Scores are finite fractional positions
+on the ordered rubric. Anthropic does not support numeric bounds in its native
+output schema, so the adapter describes them in the prompt and validates them
+after parsing. Invalid answers, refusals, and truncation fail the entire call.
+
+The adapter returns no probability distributions for Choice or Score. Boolean
+answers contain prompted estimates of P(true), validated to be finite and in
+`[0, 1]`. These estimates are not guaranteed to be calibrated. Apply thresholds
+in application code, for example `answers.requestsRefund.probability >= 0.5`.
+It respects
+`createAnthropic` configuration and forwards `providerOptions.anthropic`, including
+`structuredOutputMode`. Supported models use native structured output by default;
+`jsonTool` uses the existing Messages JSON-tool fallback. Usage, warnings,
+response information, and provider metadata are preserved.
+See [Evaluation](/docs/ai-sdk-core/evaluation).
+
+Evaluation models can also be accessed through `customProvider` aliases or
+`createProviderRegistry().evaluationModel('provider:model')`. Direct string IDs
+use Gateway by default, or an explicitly configured default provider with an
+`evaluationModel` method. See
+[model aliases and registries](/docs/ai-sdk-core/evaluation#model-aliases-and-registries).

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/anthropic",
-  "version": "4.0.55",
+  "version": "4.0.57",
   "type": "module",
   "license": "Apache-2.0",
   "sideEffects": false,
@@ -35,8 +35,8 @@
     }
   },
   "dependencies": {
-    "@ai-sdk/provider": "4.0.16",
-    "@ai-sdk/provider-utils": "5.0.42"
+    "@ai-sdk/provider": "4.0.17",
+    "@ai-sdk/provider-utils": "5.0.44"
   },
   "devDependencies": {
     "@ai-sdk/test-server": "2.0.1",

package/src/anthropic-provider.ts

@@ -2,6 +2,7 @@ import {
   InvalidArgumentError,
   NoSuchModelError,
   type Experimental_BatchV4 as BatchV4,
+  type Experimental_EvaluationModelV4 as EvaluationModelV4,
   type FilesV4,
   type LanguageModelV4,
   type ProviderV4,
@@ -16,6 +17,7 @@ import {
   withUserAgentSuffix,
   type FetchFunction,
 } from '@ai-sdk/provider-utils';
+import { Experimental_EvaluationLanguageModel as EvaluationLanguageModel } from '@ai-sdk/provider-utils/experimental-evaluation';
 import { AnthropicFiles } from './anthropic-files';
 import { AnthropicLanguageModel } from './anthropic-language-model';
 import { AnthropicBatch } from './anthropic-batch';
@@ -52,6 +54,9 @@ export interface AnthropicProvider extends ProviderV4 {
 
   messages(modelId: AnthropicModelId): LanguageModelV4;
 
+  /** Creates an experimental Choice/Score/Boolean evaluation model using Messages. */
+  evaluationModel(modelId: AnthropicModelId): EvaluationModelV4;
+
   experimental_batch(): BatchV4<{ text: AnthropicModelId }>;
 
   /**
@@ -204,6 +209,11 @@ export function createAnthropic(
   provider.languageModel = createChatModel;
   provider.chat = createChatModel;
   provider.messages = createChatModel;
+  provider.evaluationModel = (modelId: AnthropicModelId) =>
+    new EvaluationLanguageModel({
+      model: createChatModel(modelId),
+      provider: `${providerName.replace(/\.messages$/, '')}.evaluation`,
+    });
   provider.experimental_batch = createBatch;
 
   provider.embeddingModel = (modelId: string) => {