@ai-sdk/openai 2.0.0-canary.7 → 2.0.0-canary.8

package/CHANGELOG.md

@@ -1,5 +1,17 @@
 # @ai-sdk/openai
 
+## 2.0.0-canary.8
+
+### Patch Changes
+
+- 8aa9e20: feat: add speech with experimental_generateSpeech
+- Updated dependencies [5d142ab]
+- Updated dependencies [b6b43c7]
+- Updated dependencies [8aa9e20]
+- Updated dependencies [3795467]
+  - @ai-sdk/provider-utils@3.0.0-canary.8
+  - @ai-sdk/provider@2.0.0-canary.7
+
 ## 2.0.0-canary.7
 
 ### Patch Changes

package/dist/index.d.mts

@@ -1,4 +1,4 @@
-import { LanguageModelV2, ProviderV2, EmbeddingModelV2, ImageModelV1, TranscriptionModelV1 } from '@ai-sdk/provider';
+import { LanguageModelV2, ProviderV2, EmbeddingModelV2, ImageModelV1, TranscriptionModelV1, SpeechModelV1 } from '@ai-sdk/provider';
 import { FetchFunction } from '@ai-sdk/provider-utils';
 import { z } from 'zod';
 
@@ -142,6 +142,8 @@ type OpenAITranscriptionModelId = 'whisper-1' | 'gpt-4o-mini-transcribe' | 'gpt-
 
 type OpenAIResponsesModelId = 'o1' | 'o1-2024-12-17' | 'o1-mini' | 'o1-mini-2024-09-12' | 'o1-preview' | 'o1-preview-2024-09-12' | 'o3-mini' | 'o3-mini-2025-01-31' | 'gpt-4o' | 'gpt-4o-2024-05-13' | 'gpt-4o-2024-08-06' | 'gpt-4o-2024-11-20' | 'gpt-4o-mini' | 'gpt-4o-mini-2024-07-18' | 'gpt-4-turbo' | 'gpt-4-turbo-2024-04-09' | 'gpt-4-turbo-preview' | 'gpt-4-0125-preview' | 'gpt-4-1106-preview' | 'gpt-4' | 'gpt-4-0613' | 'gpt-4.5-preview' | 'gpt-4.5-preview-2025-02-27' | 'gpt-3.5-turbo-0125' | 'gpt-3.5-turbo' | 'gpt-3.5-turbo-1106' | 'chatgpt-4o-latest' | (string & {});
 
+type OpenAISpeechModelId = 'tts-1' | 'tts-1-hd' | 'gpt-4o-mini-tts' | (string & {});
+
 interface OpenAIProvider extends ProviderV2 {
     (modelId: 'gpt-3.5-turbo-instruct', settings?: OpenAICompletionSettings): OpenAICompletionLanguageModel;
     (modelId: OpenAIChatModelId, settings?: OpenAIChatSettings): LanguageModelV2;
@@ -189,6 +191,10 @@ interface OpenAIProvider extends ProviderV2 {
      */
     transcription(modelId: OpenAITranscriptionModelId): TranscriptionModelV1;
     /**
+  Creates a model for speech generation.
+     */
+    speech(modelId: OpenAISpeechModelId): SpeechModelV1;
+    /**
   OpenAI-specific tools.
      */
     tools: typeof openaiTools;

package/dist/index.d.ts

@@ -1,4 +1,4 @@
-import { LanguageModelV2, ProviderV2, EmbeddingModelV2, ImageModelV1, TranscriptionModelV1 } from '@ai-sdk/provider';
+import { LanguageModelV2, ProviderV2, EmbeddingModelV2, ImageModelV1, TranscriptionModelV1, SpeechModelV1 } from '@ai-sdk/provider';
 import { FetchFunction } from '@ai-sdk/provider-utils';
 import { z } from 'zod';
 
@@ -142,6 +142,8 @@ type OpenAITranscriptionModelId = 'whisper-1' | 'gpt-4o-mini-transcribe' | 'gpt-
 
 type OpenAIResponsesModelId = 'o1' | 'o1-2024-12-17' | 'o1-mini' | 'o1-mini-2024-09-12' | 'o1-preview' | 'o1-preview-2024-09-12' | 'o3-mini' | 'o3-mini-2025-01-31' | 'gpt-4o' | 'gpt-4o-2024-05-13' | 'gpt-4o-2024-08-06' | 'gpt-4o-2024-11-20' | 'gpt-4o-mini' | 'gpt-4o-mini-2024-07-18' | 'gpt-4-turbo' | 'gpt-4-turbo-2024-04-09' | 'gpt-4-turbo-preview' | 'gpt-4-0125-preview' | 'gpt-4-1106-preview' | 'gpt-4' | 'gpt-4-0613' | 'gpt-4.5-preview' | 'gpt-4.5-preview-2025-02-27' | 'gpt-3.5-turbo-0125' | 'gpt-3.5-turbo' | 'gpt-3.5-turbo-1106' | 'chatgpt-4o-latest' | (string & {});
 
+type OpenAISpeechModelId = 'tts-1' | 'tts-1-hd' | 'gpt-4o-mini-tts' | (string & {});
+
 interface OpenAIProvider extends ProviderV2 {
     (modelId: 'gpt-3.5-turbo-instruct', settings?: OpenAICompletionSettings): OpenAICompletionLanguageModel;
     (modelId: OpenAIChatModelId, settings?: OpenAIChatSettings): LanguageModelV2;
@@ -189,6 +191,10 @@ interface OpenAIProvider extends ProviderV2 {
      */
     transcription(modelId: OpenAITranscriptionModelId): TranscriptionModelV1;
     /**
+  Creates a model for speech generation.
+     */
+    speech(modelId: OpenAISpeechModelId): SpeechModelV1;
+    /**
   OpenAI-specific tools.
      */
     tools: typeof openaiTools;

package/dist/index.js

@@ -26,7 +26,7 @@ __export(src_exports, {
 module.exports = __toCommonJS(src_exports);
 
 // src/openai-provider.ts
-var import_provider_utils9 = require("@ai-sdk/provider-utils");
+var import_provider_utils10 = require("@ai-sdk/provider-utils");
 
 // src/openai-chat-language-model.ts
 var import_provider3 = require("@ai-sdk/provider");
@@ -560,7 +560,7 @@ var OpenAIChatLanguageModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e, _f, _g;
+    var _a, _b, _c, _d, _e, _f, _g, _h;
     const { args: body, warnings } = this.getArgs(options);
     const {
       responseHeaders,
@@ -580,10 +580,23 @@ var OpenAIChatLanguageModel = class {
       abortSignal: options.abortSignal,
       fetch: this.config.fetch
     });
-    const { messages: rawPrompt, ...rawSettings } = body;
     const choice = response.choices[0];
-    const completionTokenDetails = (_a = response.usage) == null ? void 0 : _a.completion_tokens_details;
-    const promptTokenDetails = (_b = response.usage) == null ? void 0 : _b.prompt_tokens_details;
+    const content = [];
+    const text = choice.message.content;
+    if (text != null && text.length > 0) {
+      content.push({ type: "text", text });
+    }
+    for (const toolCall of (_a = choice.message.tool_calls) != null ? _a : []) {
+      content.push({
+        type: "tool-call",
+        toolCallType: "function",
+        toolCallId: (_b = toolCall.id) != null ? _b : (0, import_provider_utils3.generateId)(),
+        toolName: toolCall.function.name,
+        args: toolCall.function.arguments
+      });
+    }
+    const completionTokenDetails = (_c = response.usage) == null ? void 0 : _c.completion_tokens_details;
+    const promptTokenDetails = (_d = response.usage) == null ? void 0 : _d.prompt_tokens_details;
     const providerMetadata = { openai: {} };
     if ((completionTokenDetails == null ? void 0 : completionTokenDetails.reasoning_tokens) != null) {
       providerMetadata.openai.reasoningTokens = completionTokenDetails == null ? void 0 : completionTokenDetails.reasoning_tokens;
@@ -598,21 +611,11 @@ var OpenAIChatLanguageModel = class {
       providerMetadata.openai.cachedPromptTokens = promptTokenDetails == null ? void 0 : promptTokenDetails.cached_tokens;
     }
     return {
-      text: choice.message.content != null ? { type: "text", text: choice.message.content } : void 0,
-      toolCalls: (_c = choice.message.tool_calls) == null ? void 0 : _c.map((toolCall) => {
-        var _a2;
-        return {
-          type: "tool-call",
-          toolCallType: "function",
-          toolCallId: (_a2 = toolCall.id) != null ? _a2 : (0, import_provider_utils3.generateId)(),
-          toolName: toolCall.function.name,
-          args: toolCall.function.arguments
-        };
-      }),
+      content,
       finishReason: mapOpenAIFinishReason(choice.finish_reason),
       usage: {
-        inputTokens: (_e = (_d = response.usage) == null ? void 0 : _d.prompt_tokens) != null ? _e : void 0,
-        outputTokens: (_g = (_f = response.usage) == null ? void 0 : _f.completion_tokens) != null ? _g : void 0
+        inputTokens: (_f = (_e = response.usage) == null ? void 0 : _e.prompt_tokens) != null ? _f : void 0,
+        outputTokens: (_h = (_g = response.usage) == null ? void 0 : _g.completion_tokens) != null ? _h : void 0
       },
       request: { body },
       response: {
@@ -660,6 +663,9 @@ var OpenAIChatLanguageModel = class {
     return {
       stream: response.pipeThrough(
         new TransformStream({
+          start(controller) {
+            controller.enqueue({ type: "stream-start", warnings });
+          },
           transform(chunk, controller) {
             var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l;
             if (!chunk.success) {
@@ -817,8 +823,7 @@ var OpenAIChatLanguageModel = class {
         })
       ),
       request: { body },
-      response: { headers: responseHeaders },
-      warnings
+      response: { headers: responseHeaders }
     };
   }
 };
@@ -1142,7 +1147,7 @@ var OpenAICompletionLanguageModel = class {
     });
     const choice = response.choices[0];
     return {
-      text: { type: "text", text: choice.text },
+      content: [{ type: "text", text: choice.text }],
       usage: {
         inputTokens: response.usage.prompt_tokens,
         outputTokens: response.usage.completion_tokens
@@ -1190,6 +1195,9 @@ var OpenAICompletionLanguageModel = class {
     return {
       stream: response.pipeThrough(
         new TransformStream({
+          start(controller) {
+            controller.enqueue({ type: "stream-start", warnings });
+          },
           transform(chunk, controller) {
             if (!chunk.success) {
               finishReason = "error";
@@ -1241,9 +1249,8 @@ var OpenAICompletionLanguageModel = class {
           }
         })
       ),
-      response: { headers: responseHeaders },
-      warnings,
-      request: { body: JSON.stringify(body) }
+      request: { body },
+      response: { headers: responseHeaders }
     };
   }
 };
@@ -1974,7 +1981,7 @@ var OpenAIResponsesLanguageModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e;
+    var _a, _b, _c, _d, _e, _f, _g, _h;
     const { args: body, warnings } = this.getArgs(options);
     const {
       responseHeaders,
@@ -2038,36 +2045,45 @@ var OpenAIResponsesLanguageModel = class {
       abortSignal: options.abortSignal,
       fetch: this.config.fetch
     });
-    const outputTextElements = response.output.filter((output) => output.type === "message").flatMap((output) => output.content).filter((content) => content.type === "output_text");
-    const toolCalls = response.output.filter((output) => output.type === "function_call").map((output) => ({
-      type: "tool-call",
-      toolCallType: "function",
-      toolCallId: output.call_id,
-      toolName: output.name,
-      args: output.arguments
-    }));
+    const content = [];
+    for (const part of response.output) {
+      switch (part.type) {
+        case "message": {
+          for (const contentPart of part.content) {
+            content.push({
+              type: "text",
+              text: contentPart.text
+            });
+            for (const annotation of contentPart.annotations) {
+              content.push({
+                type: "source",
+                sourceType: "url",
+                id: (_c = (_b = (_a = this.config).generateId) == null ? void 0 : _b.call(_a)) != null ? _c : (0, import_provider_utils8.generateId)(),
+                url: annotation.url,
+                title: annotation.title
+              });
+            }
+          }
+          break;
+        }
+        case "function_call": {
+          content.push({
+            type: "tool-call",
+            toolCallType: "function",
+            toolCallId: part.call_id,
+            toolName: part.name,
+            args: part.arguments
+          });
+          break;
+        }
+      }
+    }
     return {
-      text: {
-        type: "text",
-        text: outputTextElements.map((content) => content.text).join("\n")
-      },
-      sources: outputTextElements.flatMap(
-        (content) => content.annotations.map((annotation) => {
-          var _a2, _b2, _c2;
-          return {
-            type: "source",
-            sourceType: "url",
-            id: (_c2 = (_b2 = (_a2 = this.config).generateId) == null ? void 0 : _b2.call(_a2)) != null ? _c2 : (0, import_provider_utils8.generateId)(),
-            url: annotation.url,
-            title: annotation.title
-          };
-        })
-      ),
+      content,
       finishReason: mapOpenAIResponseFinishReason({
-        finishReason: (_a = response.incomplete_details) == null ? void 0 : _a.reason,
-        hasToolCalls: toolCalls.length > 0
+        finishReason: (_d = response.incomplete_details) == null ? void 0 : _d.reason,
+        hasToolCalls: content.some((part) => part.type === "tool-call")
       }),
-      toolCalls: toolCalls.length > 0 ? toolCalls : void 0,
       usage: {
         inputTokens: response.usage.input_tokens,
         outputTokens: response.usage.output_tokens
@@ -2083,8 +2099,8 @@ var OpenAIResponsesLanguageModel = class {
       providerMetadata: {
         openai: {
           responseId: response.id,
-          cachedPromptTokens: (_c = (_b = response.usage.input_tokens_details) == null ? void 0 : _b.cached_tokens) != null ? _c : null,
-          reasoningTokens: (_e = (_d = response.usage.output_tokens_details) == null ? void 0 : _d.reasoning_tokens) != null ? _e : null
+          cachedPromptTokens: (_f = (_e = response.usage.input_tokens_details) == null ? void 0 : _e.cached_tokens) != null ? _f : null,
+          reasoningTokens: (_h = (_g = response.usage.output_tokens_details) == null ? void 0 : _g.reasoning_tokens) != null ? _h : null
         }
       },
       warnings
@@ -2123,6 +2139,9 @@ var OpenAIResponsesLanguageModel = class {
     return {
       stream: response.pipeThrough(
         new TransformStream({
+          start(controller) {
+            controller.enqueue({ type: "stream-start", warnings });
+          },
           transform(chunk, controller) {
             var _a, _b, _c, _d, _e, _f, _g, _h;
             if (!chunk.success) {
@@ -2217,8 +2236,7 @@ var OpenAIResponsesLanguageModel = class {
         })
       ),
       request: { body },
-      response: { headers: responseHeaders },
-      warnings
+      response: { headers: responseHeaders }
     };
   }
 };
@@ -2358,14 +2376,113 @@ var openaiResponsesProviderOptionsSchema = import_zod9.z.object({
   instructions: import_zod9.z.string().nullish()
 });
 
+// src/openai-speech-model.ts
+var import_provider_utils9 = require("@ai-sdk/provider-utils");
+var import_zod10 = require("zod");
+var OpenAIProviderOptionsSchema = import_zod10.z.object({
+  instructions: import_zod10.z.string().nullish(),
+  speed: import_zod10.z.number().min(0.25).max(4).default(1).nullish()
+});
+var OpenAISpeechModel = class {
+  constructor(modelId, config) {
+    this.modelId = modelId;
+    this.config = config;
+    this.specificationVersion = "v1";
+  }
+  get provider() {
+    return this.config.provider;
+  }
+  getArgs({
+    text,
+    voice = "alloy",
+    outputFormat = "mp3",
+    speed,
+    instructions,
+    providerOptions
+  }) {
+    const warnings = [];
+    const openAIOptions = (0, import_provider_utils9.parseProviderOptions)({
+      provider: "openai",
+      providerOptions,
+      schema: OpenAIProviderOptionsSchema
+    });
+    const requestBody = {
+      model: this.modelId,
+      input: text,
+      voice,
+      response_format: "mp3",
+      speed,
+      instructions
+    };
+    if (outputFormat) {
+      if (["mp3", "opus", "aac", "flac", "wav", "pcm"].includes(outputFormat)) {
+        requestBody.response_format = outputFormat;
+      } else {
+        warnings.push({
+          type: "unsupported-setting",
+          setting: "outputFormat",
+          details: `Unsupported output format: ${outputFormat}. Using mp3 instead.`
+        });
+      }
+    }
+    if (openAIOptions) {
+      const speechModelOptions = {};
+      for (const key in speechModelOptions) {
+        const value = speechModelOptions[key];
+        if (value !== void 0) {
+          requestBody[key] = value;
+        }
+      }
+    }
+    return {
+      requestBody,
+      warnings
+    };
+  }
+  async doGenerate(options) {
+    var _a, _b, _c;
+    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, warnings } = this.getArgs(options);
+    const {
+      value: audio,
+      responseHeaders,
+      rawValue: rawResponse
+    } = await (0, import_provider_utils9.postJsonToApi)({
+      url: this.config.url({
+        path: "/audio/speech",
+        modelId: this.modelId
+      }),
+      headers: (0, import_provider_utils9.combineHeaders)(this.config.headers(), options.headers),
+      body: requestBody,
+      failedResponseHandler: openaiFailedResponseHandler,
+      successfulResponseHandler: (0, import_provider_utils9.createBinaryResponseHandler)(),
+      abortSignal: options.abortSignal,
+      fetch: this.config.fetch
+    });
+    return {
+      audio,
+      warnings,
+      request: {
+        body: JSON.stringify(requestBody)
+      },
+      response: {
+        timestamp: currentDate,
+        modelId: this.modelId,
+        headers: responseHeaders,
+        body: rawResponse
+      }
+    };
+  }
+};
+
 // src/openai-provider.ts
 function createOpenAI(options = {}) {
   var _a, _b, _c;
-  const baseURL = (_a = (0, import_provider_utils9.withoutTrailingSlash)(options.baseURL)) != null ? _a : "https://api.openai.com/v1";
+  const baseURL = (_a = (0, import_provider_utils10.withoutTrailingSlash)(options.baseURL)) != null ? _a : "https://api.openai.com/v1";
   const compatibility = (_b = options.compatibility) != null ? _b : "compatible";
   const providerName = (_c = options.name) != null ? _c : "openai";
   const getHeaders = () => ({
-    Authorization: `Bearer ${(0, import_provider_utils9.loadApiKey)({
+    Authorization: `Bearer ${(0, import_provider_utils10.loadApiKey)({
       apiKey: options.apiKey,
       environmentVariableName: "OPENAI_API_KEY",
       description: "OpenAI"
@@ -2406,6 +2523,12 @@ function createOpenAI(options = {}) {
     headers: getHeaders,
     fetch: options.fetch
   });
+  const createSpeechModel = (modelId) => new OpenAISpeechModel(modelId, {
+    provider: `${providerName}.speech`,
+    url: ({ path }) => `${baseURL}${path}`,
+    headers: getHeaders,
+    fetch: options.fetch
+  });
   const createLanguageModel = (modelId, settings) => {
     if (new.target) {
       throw new Error(
@@ -2442,6 +2565,8 @@ function createOpenAI(options = {}) {
   provider.imageModel = createImageModel;
   provider.transcription = createTranscriptionModel;
   provider.transcriptionModel = createTranscriptionModel;
+  provider.speech = createSpeechModel;
+  provider.speechModel = createSpeechModel;
   provider.tools = openaiTools;
   return provider;
 }