@ai-sdk/mistral - npm Package Compare versions

Comparing version 4.0.16 to 4.0.17

src/mistral-transcription-model-options.ts

+import {
+  lazySchema,
+  zodSchema,
+  type InferSchema,
+} from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
+
+export type MistralTranscriptionModelId = 'voxtral-mini-latest' | (string & {});
+
+// https://docs.mistral.ai/api/endpoint/audio/transcriptions
+export const mistralTranscriptionModelOptions = lazySchema(() =>
+  zodSchema(
+    z.object({
+      /**
+       * The language of the audio, e.g. "en". Providing the language can boost
+       * accuracy.
+       */
+      language: z.string().min(1).optional(),
+
+      /**
+       * The sampling temperature.
+       */
+      temperature: z.number().optional(),
+
+      /**
+       * The timestamp granularities to include in the transcription response.
+       */
+      timestampGranularities: z
+        .array(z.enum(['segment', 'word']))
+        .min(1)
+        .optional(),
+
+      /**
+       * Whether to identify speakers in the transcription.
+       */
+      diarize: z.boolean().optional(),
+
+      /**
+       * Words or phrases to guide the model toward correct spellings of names,
+       * technical terms, or domain-specific vocabulary.
+       *
+       * Mistral requires each item to omit commas and whitespace. Use
+       * underscores to represent multi-word phrases.
+       */
+      contextBias: z
+        .array(
+          z
+            .string()
+            .min(1)
+            .regex(
+              /^[^\s,]+$/,
+              'Context bias items must not contain commas or whitespace.',
+            ),
+        )
+        .max(100)
+        .optional(),
+    }),
+  ),
+);
+
+export type MistralTranscriptionModelOptions = InferSchema<
+  typeof mistralTranscriptionModelOptions
+>;

src/mistral-transcription-model.ts

+import {
+  InvalidArgumentError,
+  type JSONObject,
+  type SharedV4Warning,
+  type TranscriptionModelV4,
+} from '@ai-sdk/provider';
+import {
+  combineHeaders,
+  convertBase64ToUint8Array,
+  createJsonResponseHandler,
+  mediaTypeToExtension,
+  parseProviderOptions,
+  postFormDataToApi,
+  serializeModelOptions,
+  WORKFLOW_DESERIALIZE,
+  WORKFLOW_SERIALIZE,
+  type FetchFunction,
+} from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
+import { mistralFailedResponseHandler } from './mistral-error';
+import {
+  mistralTranscriptionModelOptions,
+  type MistralTranscriptionModelId,
+} from './mistral-transcription-model-options';
+
+type MistralTranscriptionModelConfig = {
+  provider: string;
+  baseURL: string;
+  headers?: () => Record<string, string | undefined>;
+  fetch?: FetchFunction;
+  _internal?: {
+    currentDate?: () => Date;
+  };
+};
+
+export class MistralTranscriptionModel implements TranscriptionModelV4 {
+  readonly specificationVersion = 'v4';
+
+  static [WORKFLOW_SERIALIZE](model: MistralTranscriptionModel) {
+    return serializeModelOptions({
+      modelId: model.modelId,
+      config: model.config,
+    });
+  }
+
+  static [WORKFLOW_DESERIALIZE](options: {
+    modelId: MistralTranscriptionModelId;
+    config: MistralTranscriptionModelConfig;
+  }) {
+    return new MistralTranscriptionModel(options.modelId, options.config);
+  }
+
+  get provider(): string {
+    return this.config.provider;
+  }
+
+  constructor(
+    readonly modelId: MistralTranscriptionModelId,
+    private readonly config: MistralTranscriptionModelConfig,
+  ) {}
+
+  private async getArgs({
+    audio,
+    mediaType,
+    providerOptions,
+  }: Parameters<TranscriptionModelV4['doGenerate']>[0]) {
+    const warnings: SharedV4Warning[] = [];
+    const mistralOptions = await parseProviderOptions({
+      provider: 'mistral',
+      providerOptions,
+      schema: mistralTranscriptionModelOptions,
+    });
+
+    // Mistral documents these options as mutually incompatible. Rejecting the
+    // combination locally provides a stable SDK error instead of an API 4xx.
+    if (
+      mistralOptions?.language != null &&
+      mistralOptions.timestampGranularities != null
+    ) {
+      throw new InvalidArgumentError({
+        argument: 'providerOptions',
+        message:
+          'providerOptions.mistral.language cannot be combined with providerOptions.mistral.timestampGranularities',
+      });
+    }
+
+    const formData = new FormData();
+    const blob =
+      audio instanceof Uint8Array
+        ? new Blob([audio])
+        : new Blob([convertBase64ToUint8Array(audio)]);
+
+    formData.append('model', this.modelId);
+    formData.append(
+      'file',
+      new File([blob], 'audio', { type: mediaType }),
+      `audio.${mediaTypeToExtension(mediaType)}`,
+    );
+
+    if (mistralOptions != null) {
+      appendFormValue(formData, 'language', mistralOptions.language);
+      appendFormValue(formData, 'temperature', mistralOptions.temperature);
+      appendFormValue(
+        formData,
+        'timestamp_granularities',
+        mistralOptions.timestampGranularities,
+      );
+      appendFormValue(formData, 'diarize', mistralOptions.diarize);
+      appendFormValue(formData, 'context_bias', mistralOptions.contextBias);
+    }
+
+    return { formData, warnings };
+  }
+
+  async doGenerate(
+    options: Parameters<TranscriptionModelV4['doGenerate']>[0],
+  ): Promise<Awaited<ReturnType<TranscriptionModelV4['doGenerate']>>> {
+    const currentDate = this.config._internal?.currentDate?.() ?? new Date();
+    const { formData, warnings } = await this.getArgs(options);
+
+    const {
+      value: response,
+      responseHeaders,
+      rawValue: rawResponse,
+    } = await postFormDataToApi({
+      url: `${this.config.baseURL}/audio/transcriptions`,
+      headers: combineHeaders(this.config.headers?.(), options.headers),
+      formData,
+      failedResponseHandler: mistralFailedResponseHandler,
+      successfulResponseHandler: createJsonResponseHandler(
+        mistralTranscriptionResponseSchema,
+      ),
+      abortSignal: options.abortSignal,
+      fetch: this.config.fetch,
+    });
+
+    const segments =
+      response.segments?.map(segment => ({
+        text: segment.text,
+        startSecond: segment.start,
+        endSecond: segment.end,
+      })) ?? [];
+
+    const mistralMetadata: JSONObject = {};
+
+    if (response.usage != null) {
+      mistralMetadata.usage = {
+        ...(response.usage.prompt_tokens != null && {
+          promptTokens: response.usage.prompt_tokens,
+        }),
+        ...(response.usage.completion_tokens != null && {
+          completionTokens: response.usage.completion_tokens,
+        }),
+        ...(response.usage.total_tokens != null && {
+          totalTokens: response.usage.total_tokens,
+        }),
+        ...(response.usage.prompt_audio_seconds != null && {
+          promptAudioSeconds: response.usage.prompt_audio_seconds,
+        }),
+        ...(response.usage.request_count != null && {
+          requestCount: response.usage.request_count,
+        }),
+      };
+    }
+
+    const providerSegments = response.segments
+      ?.filter(
+        segment =>
+          segment.type != null ||
+          segment.score != null ||
+          segment.speaker_id != null,
+      )
+      .map(segment => ({
+        text: segment.text,
+        startSecond: segment.start,
+        endSecond: segment.end,
+        ...(segment.type != null && { type: segment.type }),
+        ...(segment.score != null && { score: segment.score }),
+        ...(segment.speaker_id != null && {
+          speakerId: segment.speaker_id,
+        }),
+      }));
+
+    if (providerSegments != null && providerSegments.length > 0) {
+      mistralMetadata.segments = providerSegments;
+    }
+
+    return {
+      text: response.text,
+      segments,
+      language: response.language ?? undefined,
+      durationInSeconds:
+        response.usage?.prompt_audio_seconds ??
+        segments.at(-1)?.endSecond ??
+        undefined,
+      warnings,
+      response: {
+        timestamp: currentDate,
+        modelId: response.model,
+        headers: responseHeaders,
+        body: rawResponse,
+      },
+      ...(Object.keys(mistralMetadata).length > 0 && {
+        providerMetadata: {
+          mistral: mistralMetadata,
+        },
+      }),
+    };
+  }
+}
+
+function appendFormValue(
+  formData: FormData,
+  key: string,
+  value: string | number | boolean | string[] | undefined,
+): void {
+  if (value == null) {
+    return;
+  }
+
+  if (Array.isArray(value)) {
+    for (const item of value) {
+      formData.append(key, item);
+    }
+    return;
+  }
+
+  formData.append(key, String(value));
+}
+
+const mistralTranscriptionResponseSchema = z.object({
+  model: z.string(),
+  text: z.string(),
+  language: z.string().nullish(),
+  segments: z
+    .array(
+      z.object({
+        type: z.literal('transcription_segment').nullish(),
+        text: z.string(),
+        start: z.number(),
+        end: z.number(),
+        score: z.number().nullish(),
+        speaker_id: z.string().nullish(),
+      }),
+    )
+    .nullish(),
+  usage: z
+    .object({
+      prompt_tokens: z.number().nullish(),
+      completion_tokens: z.number().nullish(),
+      total_tokens: z.number().nullish(),
+      prompt_audio_seconds: z.number().nullish(),
+      request_count: z.number().nullish(),
+    })
+    .nullish(),
+});

CHANGELOG.md

 # @ai-sdk/mistral
 
+## 4.0.17
+
+### Patch Changes
+
+- 4e116c9: Add Voxtral batch transcription support through `mistral.transcription()` and `mistral.transcriptionModel()`.
+- Updated dependencies [1659cd5]
+- Updated dependencies [6a5bdff]
+  - @ai-sdk/provider-utils@5.0.15
+
 ## 4.0.16
@@ -4,0 +13,0 @@

dist/index.d.ts

@@ -1,3 +0,4 @@
-import { ProviderV4, LanguageModelV4, EmbeddingModelV4, SpeechModelV4 } from '@ai-sdk/provider';
-import { FetchFunction } from '@ai-sdk/provider-utils';
+import { ProviderV4, LanguageModelV4, EmbeddingModelV4, SpeechModelV4, TranscriptionModelV4 } from '@ai-sdk/provider';
+import * as _ai_sdk_provider_utils from '@ai-sdk/provider-utils';
+import { InferSchema, FetchFunction } from '@ai-sdk/provider-utils';
 import { z } from 'zod/v4';
@@ -40,2 +41,12 @@
 
+type MistralTranscriptionModelId = 'voxtral-mini-latest' | (string & {});
+declare const mistralTranscriptionModelOptions: _ai_sdk_provider_utils.LazySchema<{
+    language?: string | undefined;
+    temperature?: number | undefined;
+    timestampGranularities?: ("segment" | "word")[] | undefined;
+    diarize?: boolean | undefined;
+    contextBias?: string[] | undefined;
+}>;
+type MistralTranscriptionModelOptions = InferSchema<typeof mistralTranscriptionModelOptions>;
+
 interface MistralProvider extends ProviderV4 {
@@ -68,2 +79,10 @@ (modelId: MistralChatModelId): LanguageModelV4;
     /**
+     * Creates a model for audio transcription.
+     */
+    transcription(modelId: MistralTranscriptionModelId): TranscriptionModelV4;
+    /**
+     * Creates a model for audio transcription.
+     */
+    transcriptionModel(modelId: MistralTranscriptionModelId): TranscriptionModelV4;
+    /**
      * @deprecated Use `embedding` instead.
@@ -110,2 +129,2 @@ */
 
-export { type MistralEmbeddingModelOptions, type MistralLanguageModelChatOptions, type MistralLanguageModelChatOptions as MistralLanguageModelOptions, type MistralProvider, type MistralProviderSettings, type MistralSpeechModelId, type MistralSpeechModelOptions, VERSION, createMistral, mistral };
+export { type MistralEmbeddingModelOptions, type MistralLanguageModelChatOptions, type MistralLanguageModelChatOptions as MistralLanguageModelOptions, type MistralProvider, type MistralProviderSettings, type MistralSpeechModelId, type MistralSpeechModelOptions, type MistralTranscriptionModelId, type MistralTranscriptionModelOptions, VERSION, createMistral, mistral };

dist/index.js

@@ -1097,4 +1097,235 @@ // src/mistral-provider.ts
 
+// src/mistral-transcription-model.ts
+import {
+  InvalidArgumentError
+} from "@ai-sdk/provider";
+import {
+  combineHeaders as combineHeaders4,
+  convertBase64ToUint8Array,
+  createJsonResponseHandler as createJsonResponseHandler4,
+  mediaTypeToExtension,
+  parseProviderOptions as parseProviderOptions4,
+  postFormDataToApi,
+  serializeModelOptions as serializeModelOptions4,
+  WORKFLOW_DESERIALIZE as WORKFLOW_DESERIALIZE4,
+  WORKFLOW_SERIALIZE as WORKFLOW_SERIALIZE4
+} from "@ai-sdk/provider-utils";
+import { z as z9 } from "zod/v4";
+
+// src/mistral-transcription-model-options.ts
+import {
+  lazySchema,
+  zodSchema
+} from "@ai-sdk/provider-utils";
+import { z as z8 } from "zod/v4";
+var mistralTranscriptionModelOptions = lazySchema(
+  () => zodSchema(
+    z8.object({
+      /**
+       * The language of the audio, e.g. "en". Providing the language can boost
+       * accuracy.
+       */
+      language: z8.string().min(1).optional(),
+      /**
+       * The sampling temperature.
+       */
+      temperature: z8.number().optional(),
+      /**
+       * The timestamp granularities to include in the transcription response.
+       */
+      timestampGranularities: z8.array(z8.enum(["segment", "word"])).min(1).optional(),
+      /**
+       * Whether to identify speakers in the transcription.
+       */
+      diarize: z8.boolean().optional(),
+      /**
+       * Words or phrases to guide the model toward correct spellings of names,
+       * technical terms, or domain-specific vocabulary.
+       *
+       * Mistral requires each item to omit commas and whitespace. Use
+       * underscores to represent multi-word phrases.
+       */
+      contextBias: z8.array(
+        z8.string().min(1).regex(
+          /^[^\s,]+$/,
+          "Context bias items must not contain commas or whitespace."
+        )
+      ).max(100).optional()
+    })
+  )
+);
+
+// src/mistral-transcription-model.ts
+var MistralTranscriptionModel = class _MistralTranscriptionModel {
+  constructor(modelId, config) {
+    this.modelId = modelId;
+    this.config = config;
+    this.specificationVersion = "v4";
+  }
+  static [WORKFLOW_SERIALIZE4](model) {
+    return serializeModelOptions4({
+      modelId: model.modelId,
+      config: model.config
+    });
+  }
+  static [WORKFLOW_DESERIALIZE4](options) {
+    return new _MistralTranscriptionModel(options.modelId, options.config);
+  }
+  get provider() {
+    return this.config.provider;
+  }
+  async getArgs({
+    audio,
+    mediaType,
+    providerOptions
+  }) {
+    const warnings = [];
+    const mistralOptions = await parseProviderOptions4({
+      provider: "mistral",
+      providerOptions,
+      schema: mistralTranscriptionModelOptions
+    });
+    if ((mistralOptions == null ? void 0 : mistralOptions.language) != null && mistralOptions.timestampGranularities != null) {
+      throw new InvalidArgumentError({
+        argument: "providerOptions",
+        message: "providerOptions.mistral.language cannot be combined with providerOptions.mistral.timestampGranularities"
+      });
+    }
+    const formData = new FormData();
+    const blob = audio instanceof Uint8Array ? new Blob([audio]) : new Blob([convertBase64ToUint8Array(audio)]);
+    formData.append("model", this.modelId);
+    formData.append(
+      "file",
+      new File([blob], "audio", { type: mediaType }),
+      `audio.${mediaTypeToExtension(mediaType)}`
+    );
+    if (mistralOptions != null) {
+      appendFormValue(formData, "language", mistralOptions.language);
+      appendFormValue(formData, "temperature", mistralOptions.temperature);
+      appendFormValue(
+        formData,
+        "timestamp_granularities",
+        mistralOptions.timestampGranularities
+      );
+      appendFormValue(formData, "diarize", mistralOptions.diarize);
+      appendFormValue(formData, "context_bias", mistralOptions.contextBias);
+    }
+    return { formData, warnings };
+  }
+  async doGenerate(options) {
+    var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m;
+    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 { formData, warnings } = await this.getArgs(options);
+    const {
+      value: response,
+      responseHeaders,
+      rawValue: rawResponse
+    } = await postFormDataToApi({
+      url: `${this.config.baseURL}/audio/transcriptions`,
+      headers: combineHeaders4((_e = (_d = this.config).headers) == null ? void 0 : _e.call(_d), options.headers),
+      formData,
+      failedResponseHandler: mistralFailedResponseHandler,
+      successfulResponseHandler: createJsonResponseHandler4(
+        mistralTranscriptionResponseSchema
+      ),
+      abortSignal: options.abortSignal,
+      fetch: this.config.fetch
+    });
+    const segments = (_g = (_f = response.segments) == null ? void 0 : _f.map((segment) => ({
+      text: segment.text,
+      startSecond: segment.start,
+      endSecond: segment.end
+    }))) != null ? _g : [];
+    const mistralMetadata = {};
+    if (response.usage != null) {
+      mistralMetadata.usage = {
+        ...response.usage.prompt_tokens != null && {
+          promptTokens: response.usage.prompt_tokens
+        },
+        ...response.usage.completion_tokens != null && {
+          completionTokens: response.usage.completion_tokens
+        },
+        ...response.usage.total_tokens != null && {
+          totalTokens: response.usage.total_tokens
+        },
+        ...response.usage.prompt_audio_seconds != null && {
+          promptAudioSeconds: response.usage.prompt_audio_seconds
+        },
+        ...response.usage.request_count != null && {
+          requestCount: response.usage.request_count
+        }
+      };
+    }
+    const providerSegments = (_h = response.segments) == null ? void 0 : _h.filter(
+      (segment) => segment.type != null || segment.score != null || segment.speaker_id != null
+    ).map((segment) => ({
+      text: segment.text,
+      startSecond: segment.start,
+      endSecond: segment.end,
+      ...segment.type != null && { type: segment.type },
+      ...segment.score != null && { score: segment.score },
+      ...segment.speaker_id != null && {
+        speakerId: segment.speaker_id
+      }
+    }));
+    if (providerSegments != null && providerSegments.length > 0) {
+      mistralMetadata.segments = providerSegments;
+    }
+    return {
+      text: response.text,
+      segments,
+      language: (_i = response.language) != null ? _i : void 0,
+      durationInSeconds: (_m = (_l = (_j = response.usage) == null ? void 0 : _j.prompt_audio_seconds) != null ? _l : (_k = segments.at(-1)) == null ? void 0 : _k.endSecond) != null ? _m : void 0,
+      warnings,
+      response: {
+        timestamp: currentDate,
+        modelId: response.model,
+        headers: responseHeaders,
+        body: rawResponse
+      },
+      ...Object.keys(mistralMetadata).length > 0 && {
+        providerMetadata: {
+          mistral: mistralMetadata
+        }
+      }
+    };
+  }
+};
+function appendFormValue(formData, key, value) {
+  if (value == null) {
+    return;
+  }
+  if (Array.isArray(value)) {
+    for (const item of value) {
+      formData.append(key, item);
+    }
+    return;
+  }
+  formData.append(key, String(value));
+}
+var mistralTranscriptionResponseSchema = z9.object({
+  model: z9.string(),
+  text: z9.string(),
+  language: z9.string().nullish(),
+  segments: z9.array(
+    z9.object({
+      type: z9.literal("transcription_segment").nullish(),
+      text: z9.string(),
+      start: z9.number(),
+      end: z9.number(),
+      score: z9.number().nullish(),
+      speaker_id: z9.string().nullish()
+    })
+  ).nullish(),
+  usage: z9.object({
+    prompt_tokens: z9.number().nullish(),
+    completion_tokens: z9.number().nullish(),
+    total_tokens: z9.number().nullish(),
+    prompt_audio_seconds: z9.number().nullish(),
+    request_count: z9.number().nullish()
+  }).nullish()
+});
+
 // src/version.ts
-var VERSION = true ? "4.0.16" : "0.0.0-test";
+var VERSION = true ? "4.0.17" : "0.0.0-test";
 
@@ -1135,2 +1366,8 @@ // src/mistral-provider.ts
   });
+  const createTranscriptionModel = (modelId) => new MistralTranscriptionModel(modelId, {
+    provider: "mistral.transcription",
+    baseURL,
+    headers: getHeaders,
+    fetch: options.fetch
+  });
   const provider = function(modelId) {
@@ -1153,2 +1390,4 @@ if (new.target) {
   provider.speechModel = createSpeechModel;
+  provider.transcription = createTranscriptionModel;
+  provider.transcriptionModel = createTranscriptionModel;
   provider.imageModel = (modelId) => {
@@ -1155,0 +1394,0 @@ throw new NoSuchModelError({ modelId, modelType: "imageModel" });

package.json

 {
   "name": "@ai-sdk/mistral",
-  "version": "4.0.16",
+  "version": "4.0.17",
   "type": "module",
@@ -32,4 +32,4 @@ "license": "Apache-2.0",
   "dependencies": {
-    "@ai-sdk/provider": "4.0.4",
-    "@ai-sdk/provider-utils": "5.0.14"
+    "@ai-sdk/provider-utils": "5.0.15",
+    "@ai-sdk/provider": "4.0.4"
   },
@@ -36,0 +36,0 @@ "devDependencies": {

src/index.ts

@@ -16,2 +16,6 @@ export { createMistral, mistral } from './mistral-provider';
 } from './mistral-speech-model-options';
+export type {
+  MistralTranscriptionModelId,
+  MistralTranscriptionModelOptions,
+} from './mistral-transcription-model-options';
 export { VERSION } from './version';

src/mistral-provider.ts

@@ -7,2 +7,3 @@ import {
   type SpeechModelV4,
+  type TranscriptionModelV4,
 } from '@ai-sdk/provider';
@@ -21,2 +22,4 @@ import {
 import type { MistralSpeechModelId } from './mistral-speech-model-options';
+import { MistralTranscriptionModel } from './mistral-transcription-model';
+import type { MistralTranscriptionModelId } from './mistral-transcription-model-options';
 import { VERSION } from './version';
@@ -58,2 +61,14 @@
   /**
+   * Creates a model for audio transcription.
+   */
+  transcription(modelId: MistralTranscriptionModelId): TranscriptionModelV4;
+
+  /**
+   * Creates a model for audio transcription.
+   */
+  transcriptionModel(
+    modelId: MistralTranscriptionModelId,
+  ): TranscriptionModelV4;
+
+  /**
    * @deprecated Use `embedding` instead.
@@ -143,2 +158,10 @@ */
 
+  const createTranscriptionModel = (modelId: MistralTranscriptionModelId) =>
+    new MistralTranscriptionModel(modelId, {
+      provider: 'mistral.transcription',
+      baseURL,
+      headers: getHeaders,
+      fetch: options.fetch,
+    });
+
   const provider = function (modelId: MistralChatModelId) {
@@ -163,2 +186,4 @@ if (new.target) {
   provider.speechModel = createSpeechModel;
+  provider.transcription = createTranscriptionModel;
+  provider.transcriptionModel = createTranscriptionModel;
 
@@ -165,0 +190,0 @@ provider.imageModel = (modelId: string) => {

No alert changes

Improved metrics

Total package byte prevSize

263731

16.51%

Number of package files

26

8.33%

Lines of code

3337

20.51%

Worsened metrics

Number of medium supply chain risk alerts

42

27.27%

Dependency changes

• + Added

  @ai-sdk/provider-utils@5.0.15
  (transitive)

• + Added

  undici@7.29.0
  (transitive)

• - Removed

  @ai-sdk/provider-utils@5.0.14
  (transitive)

• Updated

  @ai-sdk/provider-utils@5.0.15