@ai-sdk/mistral - npm Package Compare versions

Comparing version 4.0.11 to 4.0.12

src/mistral-speech-model-options.ts

+import { z } from 'zod/v4';
+
+export type MistralSpeechModelId = 'voxtral-mini-tts-2603' | (string & {});
+
+export const mistralSpeechModelOptions = z.object({
+  /**
+   * Base64-encoded reference audio for one-off voice cloning.
+   *
+   * When provided, this takes precedence over the standard `voice` option.
+   */
+  refAudio: z.string().min(1).optional(),
+});
+
+export type MistralSpeechModelOptions = z.infer<
+  typeof mistralSpeechModelOptions
+>;

src/mistral-speech-model.ts

+import type { SharedV4Warning, SpeechModelV4 } from '@ai-sdk/provider';
+import {
+  combineHeaders,
+  createJsonResponseHandler,
+  parseProviderOptions,
+  postToApi,
+  serializeModelOptions,
+  WORKFLOW_DESERIALIZE,
+  WORKFLOW_SERIALIZE,
+  type FetchFunction,
+} from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
+import { mistralFailedResponseHandler } from './mistral-error';
+import {
+  mistralSpeechModelOptions,
+  type MistralSpeechModelId,
+} from './mistral-speech-model-options';
+
+interface MistralSpeechModelConfig {
+  provider: string;
+  baseURL: string;
+  headers?: () => Record<string, string | undefined>;
+  fetch?: FetchFunction;
+  _internal?: {
+    currentDate?: () => Date;
+  };
+}
+
+type MistralSpeechOutputFormat = 'pcm' | 'wav' | 'mp3' | 'flac' | 'opus';
+
+export class MistralSpeechModel implements SpeechModelV4 {
+  readonly specificationVersion = 'v4';
+
+  static [WORKFLOW_SERIALIZE](model: MistralSpeechModel) {
+    return serializeModelOptions({
+      modelId: model.modelId,
+      config: model.config,
+    });
+  }
+
+  static [WORKFLOW_DESERIALIZE](options: {
+    modelId: MistralSpeechModelId;
+    config: MistralSpeechModelConfig;
+  }) {
+    return new MistralSpeechModel(options.modelId, options.config);
+  }
+
+  get provider(): string {
+    return this.config.provider;
+  }
+
+  constructor(
+    readonly modelId: MistralSpeechModelId,
+    private readonly config: MistralSpeechModelConfig,
+  ) {}
+
+  private async getArgs({
+    text,
+    voice,
+    outputFormat = 'mp3',
+    instructions,
+    speed,
+    language,
+    providerOptions,
+  }: Parameters<SpeechModelV4['doGenerate']>[0]) {
+    const warnings: SharedV4Warning[] = [];
+    const mistralOptions = await parseProviderOptions({
+      provider: 'mistral',
+      providerOptions,
+      schema: mistralSpeechModelOptions,
+    });
+
+    let responseFormat: MistralSpeechOutputFormat = 'mp3';
+    if (['pcm', 'wav', 'mp3', 'flac', 'opus'].includes(outputFormat)) {
+      responseFormat = outputFormat as MistralSpeechOutputFormat;
+    } else {
+      warnings.push({
+        type: 'unsupported',
+        feature: 'outputFormat',
+        details: `Unsupported output format: ${outputFormat}. Using mp3 instead.`,
+      });
+    }
+
+    if (instructions != null) {
+      warnings.push({
+        type: 'unsupported',
+        feature: 'instructions',
+        details:
+          'Mistral speech models do not support the `instructions` option. ' +
+          'Use a reference audio clip to guide delivery.',
+      });
+    }
+
+    if (speed != null) {
+      warnings.push({
+        type: 'unsupported',
+        feature: 'speed',
+        details:
+          'Mistral speech models do not support the `speed` option. It was ignored.',
+      });
+    }
+
+    if (language != null) {
+      warnings.push({
+        type: 'unsupported',
+        feature: 'language',
+        details:
+          'Mistral speech models do not support the `language` option. ' +
+          'Language is inferred from the input text and voice.',
+      });
+    }
+
+    const refAudio = mistralOptions?.refAudio;
+    const requestBody = {
+      model: this.modelId,
+      input: text,
+      voice_id: refAudio == null ? voice : undefined,
+      ref_audio: refAudio,
+      response_format: responseFormat,
+      stream: false,
+    };
+
+    const requestBodyValues = {
+      ...requestBody,
+      ref_audio: refAudio == null ? undefined : '[redacted]',
+    };
+
+    return { requestBody, requestBodyValues, warnings };
+  }
+
+  async doGenerate(
+    options: Parameters<SpeechModelV4['doGenerate']>[0],
+  ): Promise<Awaited<ReturnType<SpeechModelV4['doGenerate']>>> {
+    const currentDate = this.config._internal?.currentDate?.() ?? new Date();
+    const { requestBody, requestBodyValues, warnings } =
+      await this.getArgs(options);
+
+    const {
+      value: response,
+      responseHeaders,
+      rawValue: rawResponse,
+    } = await postToApi({
+      url: `${this.config.baseURL}/audio/speech`,
+      headers: combineHeaders(
+        { 'Content-Type': 'application/json' },
+        this.config.headers?.(),
+        options.headers,
+      ),
+      body: {
+        content: JSON.stringify(requestBody),
+        values: requestBodyValues,
+      },
+      failedResponseHandler: mistralFailedResponseHandler,
+      successfulResponseHandler: createJsonResponseHandler(
+        mistralSpeechResponseSchema,
+      ),
+      abortSignal: options.abortSignal,
+      fetch: this.config.fetch,
+    });
+
+    return {
+      audio: response.audio_data,
+      warnings,
+      request: {
+        body: JSON.stringify(requestBodyValues),
+      },
+      response: {
+        timestamp: currentDate,
+        modelId: this.modelId,
+        headers: responseHeaders,
+        body: rawResponse,
+      },
+    };
+  }
+}
+
+const mistralSpeechResponseSchema = z.object({
+  audio_data: z.string(),
+});

CHANGELOG.md

 # @ai-sdk/mistral
 
+## 4.0.12
+
+### Patch Changes
+
+- ba433f7: Add non-streaming Voxtral text-to-speech generation with saved voice IDs and one-off reference audio.
+
 ## 4.0.11
@@ -4,0 +10,0 @@

dist/index.d.ts

@@ -1,2 +0,2 @@
-import { ProviderV4, LanguageModelV4, EmbeddingModelV4 } from '@ai-sdk/provider';
+import { ProviderV4, LanguageModelV4, EmbeddingModelV4, SpeechModelV4 } from '@ai-sdk/provider';
 import { FetchFunction } from '@ai-sdk/provider-utils';
@@ -34,2 +34,8 @@ import { z } from 'zod/v4';
 
+type MistralSpeechModelId = 'voxtral-mini-tts-2603' | (string & {});
+declare const mistralSpeechModelOptions: z.ZodObject<{
+    refAudio: z.ZodOptional<z.ZodString>;
+}, z.core.$strip>;
+type MistralSpeechModelOptions = z.infer<typeof mistralSpeechModelOptions>;
+
 interface MistralProvider extends ProviderV4 {
@@ -54,2 +60,10 @@ (modelId: MistralChatModelId): LanguageModelV4;
     /**
+     * Creates a model for speech generation (text-to-speech).
+     */
+    speech(modelId: MistralSpeechModelId): SpeechModelV4;
+    /**
+     * Creates a model for speech generation (text-to-speech).
+     */
+    speechModel(modelId: MistralSpeechModelId): SpeechModelV4;
+    /**
      * @deprecated Use `embedding` instead.
@@ -96,2 +110,2 @@ */
 
-export { type MistralEmbeddingModelOptions, type MistralLanguageModelChatOptions, type MistralLanguageModelChatOptions as MistralLanguageModelOptions, type MistralProvider, type MistralProviderSettings, VERSION, createMistral, mistral };
+export { type MistralEmbeddingModelOptions, type MistralLanguageModelChatOptions, type MistralLanguageModelChatOptions as MistralLanguageModelOptions, type MistralProvider, type MistralProviderSettings, type MistralSpeechModelId, type MistralSpeechModelOptions, VERSION, createMistral, mistral };

dist/index.js

@@ -941,4 +941,152 @@ // src/mistral-provider.ts
 
+// src/mistral-speech-model.ts
+import {
+  combineHeaders as combineHeaders3,
+  createJsonResponseHandler as createJsonResponseHandler3,
+  parseProviderOptions as parseProviderOptions3,
+  postToApi,
+  serializeModelOptions as serializeModelOptions3,
+  WORKFLOW_DESERIALIZE as WORKFLOW_DESERIALIZE3,
+  WORKFLOW_SERIALIZE as WORKFLOW_SERIALIZE3
+} from "@ai-sdk/provider-utils";
+import { z as z7 } from "zod/v4";
+
+// src/mistral-speech-model-options.ts
+import { z as z6 } from "zod/v4";
+var mistralSpeechModelOptions = z6.object({
+  /**
+   * Base64-encoded reference audio for one-off voice cloning.
+   *
+   * When provided, this takes precedence over the standard `voice` option.
+   */
+  refAudio: z6.string().min(1).optional()
+});
+
+// src/mistral-speech-model.ts
+var MistralSpeechModel = class _MistralSpeechModel {
+  constructor(modelId, config) {
+    this.modelId = modelId;
+    this.config = config;
+    this.specificationVersion = "v4";
+  }
+  static [WORKFLOW_SERIALIZE3](model) {
+    return serializeModelOptions3({
+      modelId: model.modelId,
+      config: model.config
+    });
+  }
+  static [WORKFLOW_DESERIALIZE3](options) {
+    return new _MistralSpeechModel(options.modelId, options.config);
+  }
+  get provider() {
+    return this.config.provider;
+  }
+  async getArgs({
+    text,
+    voice,
+    outputFormat = "mp3",
+    instructions,
+    speed,
+    language,
+    providerOptions
+  }) {
+    const warnings = [];
+    const mistralOptions = await parseProviderOptions3({
+      provider: "mistral",
+      providerOptions,
+      schema: mistralSpeechModelOptions
+    });
+    let responseFormat = "mp3";
+    if (["pcm", "wav", "mp3", "flac", "opus"].includes(outputFormat)) {
+      responseFormat = outputFormat;
+    } else {
+      warnings.push({
+        type: "unsupported",
+        feature: "outputFormat",
+        details: `Unsupported output format: ${outputFormat}. Using mp3 instead.`
+      });
+    }
+    if (instructions != null) {
+      warnings.push({
+        type: "unsupported",
+        feature: "instructions",
+        details: "Mistral speech models do not support the `instructions` option. Use a reference audio clip to guide delivery."
+      });
+    }
+    if (speed != null) {
+      warnings.push({
+        type: "unsupported",
+        feature: "speed",
+        details: "Mistral speech models do not support the `speed` option. It was ignored."
+      });
+    }
+    if (language != null) {
+      warnings.push({
+        type: "unsupported",
+        feature: "language",
+        details: "Mistral speech models do not support the `language` option. Language is inferred from the input text and voice."
+      });
+    }
+    const refAudio = mistralOptions == null ? void 0 : mistralOptions.refAudio;
+    const requestBody = {
+      model: this.modelId,
+      input: text,
+      voice_id: refAudio == null ? voice : void 0,
+      ref_audio: refAudio,
+      response_format: responseFormat,
+      stream: false
+    };
+    const requestBodyValues = {
+      ...requestBody,
+      ref_audio: refAudio == null ? void 0 : "[redacted]"
+    };
+    return { requestBody, requestBodyValues, warnings };
+  }
+  async doGenerate(options) {
+    var _a, _b, _c, _d, _e;
+    const currentDate = (_c = (_b = (_a = this.config._internal) == null ? void 0 : _a.currentDate) == null ? void 0 : _b.call(_a)) != null ? _c : /* @__PURE__ */ new Date();
+    const { requestBody, requestBodyValues, warnings } = await this.getArgs(options);
+    const {
+      value: response,
+      responseHeaders,
+      rawValue: rawResponse
+    } = await postToApi({
+      url: `${this.config.baseURL}/audio/speech`,
+      headers: combineHeaders3(
+        { "Content-Type": "application/json" },
+        (_e = (_d = this.config).headers) == null ? void 0 : _e.call(_d),
+        options.headers
+      ),
+      body: {
+        content: JSON.stringify(requestBody),
+        values: requestBodyValues
+      },
+      failedResponseHandler: mistralFailedResponseHandler,
+      successfulResponseHandler: createJsonResponseHandler3(
+        mistralSpeechResponseSchema
+      ),
+      abortSignal: options.abortSignal,
+      fetch: this.config.fetch
+    });
+    return {
+      audio: response.audio_data,
+      warnings,
+      request: {
+        body: JSON.stringify(requestBodyValues)
+      },
+      response: {
+        timestamp: currentDate,
+        modelId: this.modelId,
+        headers: responseHeaders,
+        body: rawResponse
+      }
+    };
+  }
+};
+var mistralSpeechResponseSchema = z7.object({
+  audio_data: z7.string()
+});
+
 // src/version.ts
-var VERSION = true ? "4.0.11" : "0.0.0-test";
+var VERSION = true ? "4.0.12" : "0.0.0-test";
 
@@ -973,2 +1121,8 @@ // src/mistral-provider.ts
   });
+  const createSpeechModel = (modelId) => new MistralSpeechModel(modelId, {
+    provider: "mistral.speech",
+    baseURL,
+    headers: getHeaders,
+    fetch: options.fetch
+  });
   const provider = function(modelId) {
@@ -989,2 +1143,4 @@ if (new.target) {
   provider.textEmbeddingModel = createEmbeddingModel;
+  provider.speech = createSpeechModel;
+  provider.speechModel = createSpeechModel;
   provider.imageModel = (modelId) => {
@@ -991,0 +1147,0 @@ throw new NoSuchModelError({ modelId, modelType: "imageModel" });

package.json

 {
   "name": "@ai-sdk/mistral",
-  "version": "4.0.11",
+  "version": "4.0.12",
   "type": "module",
@@ -5,0 +5,0 @@ "license": "Apache-2.0",

README.md

 # AI SDK - Mistral Provider
 
-The **[Mistral provider](https://ai-sdk.dev/providers/ai-sdk-providers/mistral)** for the [AI SDK](https://ai-sdk.dev/docs) contains language model support for the Mistral chat API.
+The **[Mistral provider](https://ai-sdk.dev/providers/ai-sdk-providers/mistral)** for the [AI SDK](https://ai-sdk.dev/docs) contains language model, embedding model, and speech model support for Mistral APIs.
 
@@ -5,0 +5,0 @@ > **Deploying to Vercel?** With Vercel's AI Gateway you can access Mistral (and hundreds of models from other providers) — no additional packages, API keys, or extra cost. [Get started with AI Gateway](https://vercel.com/ai-gateway).

src/index.ts

@@ -12,2 +12,6 @@ export { createMistral, mistral } from './mistral-provider';
 export type { MistralEmbeddingModelOptions } from './mistral-embedding-model-options';
+export type {
+  MistralSpeechModelId,
+  MistralSpeechModelOptions,
+} from './mistral-speech-model-options';
 export { VERSION } from './version';

src/mistral-provider.ts

@@ -6,2 +6,3 @@ import {
   type ProviderV4,
+  type SpeechModelV4,
 } from '@ai-sdk/provider';
@@ -18,2 +19,4 @@ import {
 import type { MistralEmbeddingModelId } from './mistral-embedding-model-options';
+import { MistralSpeechModel } from './mistral-speech-model';
+import type { MistralSpeechModelId } from './mistral-speech-model-options';
 import { VERSION } from './version';
@@ -45,2 +48,12 @@
   /**
+   * Creates a model for speech generation (text-to-speech).
+   */
+  speech(modelId: MistralSpeechModelId): SpeechModelV4;
+
+  /**
+   * Creates a model for speech generation (text-to-speech).
+   */
+  speechModel(modelId: MistralSpeechModelId): SpeechModelV4;
+
+  /**
    * @deprecated Use `embedding` instead.
@@ -122,2 +135,10 @@ */
 
+  const createSpeechModel = (modelId: MistralSpeechModelId) =>
+    new MistralSpeechModel(modelId, {
+      provider: 'mistral.speech',
+      baseURL,
+      headers: getHeaders,
+      fetch: options.fetch,
+    });
+
   const provider = function (modelId: MistralChatModelId) {
@@ -140,2 +161,4 @@ if (new.target) {
   provider.textEmbeddingModel = createEmbeddingModel;
+  provider.speech = createSpeechModel;
+  provider.speechModel = createSpeechModel;
 
@@ -142,0 +165,0 @@ provider.imageModel = (modelId: string) => {

No alert changes

Improved metrics

Total package byte prevSize

224142

11.55%

Number of package files

24

9.09%

Lines of code

2743

15.3%

Worsened metrics

Number of medium supply chain risk alerts

33

37.5%

No dependency changes