feat(mistral): added magistral reasoning models (vercel#6715)

## background

Mistral released two new reasoning models (`magistral-small-2506` and
`magistral-medium-2506`) that use `<think>...</think>` tags to separate
reasoning from final responses.

## summary

- add extract reasoning middleware examples with mistral
- add model ids - magistral-small-2506, magistral-medium-2506

## verification

- all tests pass including new reasoning-specific test with middleware 
- content separation works in both `generateText` and `streamText` modes
- examples demonstrate magistral reasoning using extract reasoning
middleware

## future work
* update provider to extract reasoning once mistral api exposes native
reasoning events

---------

Co-authored-by: Josh [email protected]

Author: dancer

.changeset/lemon-yaks-move.md

@@ -0,0 +1,5 @@
+---
+'@ai-sdk/mistral': patch
+---
+
+feat(mistral): added magistral reasoning models

content/providers/01-ai-sdk-providers/20-mistral.mdx

@@ -143,6 +143,42 @@ const result = await generateText({
 });
 ```
 
+### Reasoning Models
+
+Mistral offers reasoning models that provide step-by-step thinking capabilities:
+
+- **magistral-small-2506**: Smaller reasoning model for efficient step-by-step thinking
+- **magistral-medium-2506**: More powerful reasoning model balancing performance and cost
+
+These models return content that includes `<think>...</think>` tags containing the reasoning process. To properly extract and separate the reasoning from the final answer, use the [extract reasoning middleware](/docs/reference/ai-sdk-core/extract-reasoning-middleware):
+
+```ts
+import { mistral } from '@ai-sdk/mistral';
+import {
+  extractReasoningMiddleware,
+  generateText,
+  wrapLanguageModel,
+} from 'ai';
+
+const result = await generateText({
+  model: wrapLanguageModel({
+    model: mistral('magistral-small-2506'),
+    middleware: extractReasoningMiddleware({
+      tagName: 'think',
+    }),
+  }),
+  prompt: 'What is 15 * 24?',
+});
+
+console.log('REASONING:', result.reasoningText);
+// Output: "Let me calculate this step by step..."
+
+console.log('ANSWER:', result.text);
+// Output: "360"
+```
+
+The middleware automatically parses the `<think>` tags and provides separate `reasoningText` and `text` properties in the result.
+
 ### Example
 
 You can use Mistral language models to generate text with the `generateText` function:
@@ -162,14 +198,16 @@ Mistral language models can also be used in the `streamText`, `generateObject`,
 
 ### Model Capabilities
 
-| Model                  | Image Input         | Object Generation   | Tool Usage          | Tool Streaming      |
-| ---------------------- | ------------------- | ------------------- | ------------------- | ------------------- |
-| `pixtral-large-latest` | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
-| `mistral-large-latest` | <Cross size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
-| `mistral-small-latest` | <Cross size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
-| `ministral-3b-latest`  | <Cross size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
-| `ministral-8b-latest`  | <Cross size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
-| `pixtral-12b-2409`     | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
+| Model                   | Image Input         | Object Generation   | Tool Usage          | Tool Streaming      |
+| ----------------------- | ------------------- | ------------------- | ------------------- | ------------------- |
+| `pixtral-large-latest`  | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
+| `mistral-large-latest`  | <Cross size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
+| `mistral-small-latest`  | <Cross size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
+| `magistral-small-2506`  | <Cross size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
+| `magistral-medium-2506` | <Cross size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
+| `ministral-3b-latest`   | <Cross size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
+| `ministral-8b-latest`   | <Cross size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
+| `pixtral-12b-2409`      | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> | <Check size={18} /> |
 
 <Note>
   The table above lists popular models. Please see the [Mistral

examples/ai-core/src/generate-text/mistral-reasoning-raw.ts

@@ -0,0 +1,32 @@
+import { mistral } from '@ai-sdk/mistral';
+import {
+  extractReasoningMiddleware,
+  generateText,
+  wrapLanguageModel,
+} from 'ai';
+import 'dotenv/config';
+
+async function main() {
+  const result = await generateText({
+    model: wrapLanguageModel({
+      model: mistral('magistral-medium-2506'),
+      middleware: extractReasoningMiddleware({
+        tagName: 'think',
+      }),
+    }),
+    prompt:
+      'Solve this step by step: If a train travels 60 mph for 2 hours, how far does it go?',
+    maxOutputTokens: 500,
+  });
+
+  console.log('\nREASONING:\n');
+  console.log(result.reasoningText);
+
+  console.log('\nTEXT:\n');
+  console.log(result.text);
+
+  console.log();
+  console.log('Usage:', result.usage);
+}
+
+main().catch(console.error);

examples/ai-core/src/stream-text/mistral-reasoning-raw.ts

@@ -0,0 +1,43 @@
+import { mistral } from '@ai-sdk/mistral';
+import { extractReasoningMiddleware, streamText, wrapLanguageModel } from 'ai';
+import 'dotenv/config';
+
+async function main() {
+  const result = streamText({
+    model: wrapLanguageModel({
+      model: mistral('magistral-small-2506'),
+      middleware: extractReasoningMiddleware({
+        tagName: 'think',
+      }),
+    }),
+    prompt: 'What is 2 + 2?',
+  });
+
+  console.log('Mistral reasoning model with extracted reasoning:');
+  console.log();
+
+  let enteredReasoning = false;
+  let enteredText = false;
+
+  for await (const part of result.fullStream) {
+    if (part.type === 'reasoning') {
+      if (!enteredReasoning) {
+        enteredReasoning = true;
+        console.log('REASONING:');
+      }
+      process.stdout.write(part.text);
+    } else if (part.type === 'text') {
+      if (!enteredText) {
+        enteredText = true;
+        console.log('\n\nTEXT:');
+      }
+      process.stdout.write(part.text);
+    }
+  }
+
+  console.log();
+  console.log();
+  console.log('Usage:', await result.usage);
+}
+
+main().catch(console.error);

packages/mistral/src/mistral-chat-language-model.test.ts

@@ -377,6 +377,34 @@ describe('doGenerate', () => {
       ]
     `);
   });
+
+  it('should return raw text with think tags for reasoning models', async () => {
+    const reasoningModel = provider.chat('magistral-small-2506');
+
+    prepareJsonResponse({
+      content:
+        "<think>\nLet me think about this problem step by step.\nFirst, I need to understand what the user is asking.\nThen I can provide a helpful response.\n</think>\n\nHello! I'm ready to help you with your question.",
+    });
+
+    const { content } = await reasoningModel.doGenerate({
+      prompt: TEST_PROMPT,
+    });
+
+    expect(content).toMatchInlineSnapshot(`
+      [
+        {
+          "text": "<think>
+      Let me think about this problem step by step.
+      First, I need to understand what the user is asking.
+      Then I can provide a helpful response.
+      </think>
+
+      Hello! I'm ready to help you with your question.",
+          "type": "text",
+        },
+      ]
+    `);
+  });
 });
 
 describe('doStream', () => {
@@ -772,9 +800,7 @@ describe('doStream with raw chunks', () => {
       includeRawChunks: true,
     });
 
-    const chunks = await convertReadableStreamToArray(stream);
-
-    expect(chunks).toMatchInlineSnapshot(`
+    expect(await convertReadableStreamToArray(stream)).toMatchInlineSnapshot(`
       [
         {
           "type": "stream-start",

packages/mistral/src/mistral-chat-options.ts

@@ -8,6 +8,9 @@ export type MistralChatModelId =
   | 'mistral-large-latest'
   | 'mistral-small-latest'
   | 'pixtral-large-latest'
+  // reasoning models
+  | 'magistral-small-2506'
+  | 'magistral-medium-2506'
   // free
   | 'pixtral-12b-2409'
   // legacy