@ai-sdk/alibaba 2.0.0-beta.9 → 2.0.0-canary.39

package/src/alibaba-chat-language-model.ts

@@ -3,34 +3,39 @@ import {
   mapOpenAICompatibleFinishReason,
   prepareTools,
 } from '@ai-sdk/openai-compatible/internal';
-import {
-  InvalidResponseDataError,
-  type LanguageModelV4,
-  type LanguageModelV4CallOptions,
-  type LanguageModelV4Content,
-  type LanguageModelV4FinishReason,
-  type LanguageModelV4GenerateResult,
-  type LanguageModelV4StreamPart,
-  type LanguageModelV4StreamResult,
-  type SharedV4Warning,
+import type {
+  LanguageModelV4,
+  LanguageModelV4CallOptions,
+  LanguageModelV4Content,
+  LanguageModelV4FinishReason,
+  LanguageModelV4GenerateResult,
+  LanguageModelV4StreamPart,
+  LanguageModelV4StreamResult,
+  SharedV4Warning,
 } from '@ai-sdk/provider';
 import {
   combineHeaders,
   createEventSourceResponseHandler,
   createJsonResponseHandler,
   generateId,
-  isParsableJson,
+  isCustomReasoning,
+  mapReasoningToProviderBudget,
   parseProviderOptions,
   postJsonToApi,
+  serializeModelOptions,
+  StreamingToolCallTracker,
+  WORKFLOW_SERIALIZE,
+  WORKFLOW_DESERIALIZE,
+  type InferSchema,
   type ParseResult,
 } from '@ai-sdk/provider-utils';
 import { z } from 'zod/v4';
 import {
-  alibabaLanguageModelOptions,
+  alibabaLanguageModelChatOptions,
   type AlibabaChatModelId,
-} from './alibaba-chat-options';
+} from './alibaba-chat-language-model-options';
 import type { AlibabaConfig } from './alibaba-config';
-import { alibabaFailedResponseHandler } from './alibaba-provider';
+import { alibabaFailedResponseHandler } from './alibaba-error';
 import { convertAlibabaUsage } from './convert-alibaba-usage';
 import { convertToAlibabaChatMessages } from './convert-to-alibaba-chat-messages';
 import { CacheControlValidator } from './get-cache-control';
@@ -44,12 +49,26 @@ import { CacheControlValidator } from './get-cache-control';
  * - Thinking budget control (thinking_budget)
  * - Prompt caching (cached_tokens tracking)
  */
-export class AlibabaLanguageModel implements LanguageModelV4 {
+export class AlibabaChatLanguageModel implements LanguageModelV4 {
   readonly specificationVersion = 'v4';
   readonly modelId: AlibabaChatModelId;
 
   private readonly config: AlibabaConfig;
 
+  static [WORKFLOW_SERIALIZE](model: AlibabaChatLanguageModel) {
+    return serializeModelOptions({
+      modelId: model.modelId,
+      config: model.config,
+    });
+  }
+
+  static [WORKFLOW_DESERIALIZE](options: {
+    modelId: AlibabaChatModelId;
+    config: AlibabaConfig;
+  }) {
+    return new AlibabaChatLanguageModel(options.modelId, options.config);
+  }
+
   constructor(modelId: AlibabaChatModelId, config: AlibabaConfig) {
     this.modelId = modelId;
     this.config = config;
@@ -78,6 +97,7 @@ export class AlibabaLanguageModel implements LanguageModelV4 {
     stopSequences,
     responseFormat,
     seed,
+    reasoning,
     providerOptions,
     tools,
     toolChoice,
@@ -89,7 +109,7 @@ export class AlibabaLanguageModel implements LanguageModelV4 {
     const alibabaOptions = await parseProviderOptions({
       provider: 'alibaba',
       providerOptions,
-      schema: alibabaLanguageModelOptions,
+      schema: alibabaLanguageModelChatOptions,
     });
 
     // Warn about unsupported features
@@ -121,13 +141,11 @@ export class AlibabaLanguageModel implements LanguageModelV4 {
             : { type: 'json_object' }
           : undefined,
 
-      // Alibaba-specific options
-      ...(alibabaOptions?.enableThinking != null
-        ? { enable_thinking: alibabaOptions.enableThinking }
-        : {}),
-      ...(alibabaOptions?.thinkingBudget != null
-        ? { thinking_budget: alibabaOptions.thinkingBudget }
-        : {}),
+      ...resolveAlibabaThinking({
+        reasoning,
+        alibabaOptions,
+        warnings,
+      }),
 
       // Convert messages with cache control support
       messages: convertToAlibabaChatMessages({
@@ -170,7 +188,7 @@ export class AlibabaLanguageModel implements LanguageModelV4 {
       rawValue: rawResponse,
     } = await postJsonToApi({
       url: `${this.config.baseURL}/chat/completions`,
-      headers: combineHeaders(this.config.headers(), options.headers),
+      headers: combineHeaders(this.config.headers?.(), options.headers),
       body: args,
       failedResponseHandler: alibabaFailedResponseHandler,
       successfulResponseHandler: createJsonResponseHandler(
@@ -203,7 +221,7 @@ export class AlibabaLanguageModel implements LanguageModelV4 {
       for (const toolCall of choice.message.tool_calls) {
         content.push({
           type: 'tool-call',
-          toolCallId: toolCall.id,
+          toolCallId: toolCall.id ?? generateId(),
           toolName: toolCall.function.name,
           input: toolCall.function.arguments!,
         });
@@ -241,7 +259,7 @@ export class AlibabaLanguageModel implements LanguageModelV4 {
 
     const { responseHeaders, value: response } = await postJsonToApi({
       url: `${this.config.baseURL}/chat/completions`,
-      headers: combineHeaders(this.config.headers(), options.headers),
+      headers: combineHeaders(this.config.headers?.(), options.headers),
       body,
       failedResponseHandler: alibabaFailedResponseHandler,
       successfulResponseHandler: createEventSourceResponseHandler(
@@ -262,13 +280,7 @@ export class AlibabaLanguageModel implements LanguageModelV4 {
     let activeText = false;
     let activeReasoningId: string | null = null;
 
-    // Track tool calls for accumulation across chunks
-    const toolCalls: Array<{
-      id: string;
-      type: 'function';
-      function: { name: string; arguments: string };
-      hasFinished: boolean;
-    }> = [];
+    let toolCallTracker: StreamingToolCallTracker;
 
     return {
       stream: response.pipeThrough(
@@ -277,6 +289,9 @@ export class AlibabaLanguageModel implements LanguageModelV4 {
           LanguageModelV4StreamPart
         >({
           start(controller) {
+            toolCallTracker = new StreamingToolCallTracker(controller, {
+              generateId,
+            });
             controller.enqueue({ type: 'stream-start', warnings });
           },
 
@@ -381,106 +396,7 @@ export class AlibabaLanguageModel implements LanguageModelV4 {
               }
 
               for (const toolCallDelta of delta.tool_calls) {
-                const index = toolCallDelta.index ?? toolCalls.length;
-
-                // New tool call - first chunk with id and name
-                if (toolCalls[index] == null) {
-                  if (toolCallDelta.id == null) {
-                    throw new InvalidResponseDataError({
-                      data: toolCallDelta,
-                      message: `Expected 'id' to be a string.`,
-                    });
-                  }
-
-                  if (toolCallDelta.function?.name == null) {
-                    throw new InvalidResponseDataError({
-                      data: toolCallDelta,
-                      message: `Expected 'function.name' to be a string.`,
-                    });
-                  }
-
-                  controller.enqueue({
-                    type: 'tool-input-start',
-                    id: toolCallDelta.id,
-                    toolName: toolCallDelta.function.name,
-                  });
-
-                  toolCalls[index] = {
-                    id: toolCallDelta.id,
-                    type: 'function',
-                    function: {
-                      name: toolCallDelta.function.name,
-                      arguments: toolCallDelta.function.arguments ?? '',
-                    },
-                    hasFinished: false,
-                  };
-
-                  const toolCall = toolCalls[index];
-
-                  // Send initial delta if arguments started
-                  if (toolCall.function.arguments.length > 0) {
-                    controller.enqueue({
-                      type: 'tool-input-delta',
-                      id: toolCall.id,
-                      delta: toolCall.function.arguments,
-                    });
-                  }
-
-                  // Check if already complete (some providers send full tool call at once)
-                  if (isParsableJson(toolCall.function.arguments)) {
-                    controller.enqueue({
-                      type: 'tool-input-end',
-                      id: toolCall.id,
-                    });
-
-                    controller.enqueue({
-                      type: 'tool-call',
-                      toolCallId: toolCall.id,
-                      toolName: toolCall.function.name,
-                      input: toolCall.function.arguments,
-                    });
-
-                    toolCall.hasFinished = true;
-                  }
-
-                  continue;
-                }
-
-                // Existing tool call - accumulate arguments
-                const toolCall = toolCalls[index];
-
-                if (toolCall.hasFinished) {
-                  continue;
-                }
-
-                // Append arguments if not null (skip arguments: null chunks)
-                if (toolCallDelta.function?.arguments != null) {
-                  toolCall.function.arguments +=
-                    toolCallDelta.function.arguments;
-
-                  controller.enqueue({
-                    type: 'tool-input-delta',
-                    id: toolCall.id,
-                    delta: toolCallDelta.function.arguments,
-                  });
-                }
-
-                // Check if tool call is now complete
-                if (isParsableJson(toolCall.function.arguments)) {
-                  controller.enqueue({
-                    type: 'tool-input-end',
-                    id: toolCall.id,
-                  });
-
-                  controller.enqueue({
-                    type: 'tool-call',
-                    toolCallId: toolCall.id,
-                    toolName: toolCall.function.name,
-                    input: toolCall.function.arguments,
-                  });
-
-                  toolCall.hasFinished = true;
-                }
+                toolCallTracker.processDelta(toolCallDelta);
               }
             }
 
@@ -505,6 +421,8 @@ export class AlibabaLanguageModel implements LanguageModelV4 {
               controller.enqueue({ type: 'text-end', id: '0' });
             }
 
+            toolCallTracker.flush();
+
             controller.enqueue({
               type: 'finish',
               finishReason,
@@ -519,6 +437,52 @@ export class AlibabaLanguageModel implements LanguageModelV4 {
   }
 }
 
+function resolveAlibabaThinking({
+  reasoning,
+  alibabaOptions,
+  warnings,
+}: {
+  reasoning: LanguageModelV4CallOptions['reasoning'];
+  alibabaOptions:
+    | InferSchema<typeof alibabaLanguageModelChatOptions>
+    | undefined;
+  warnings: SharedV4Warning[];
+}): { enable_thinking?: boolean; thinking_budget?: number } {
+  if (
+    alibabaOptions?.enableThinking != null ||
+    alibabaOptions?.thinkingBudget != null
+  ) {
+    return {
+      ...(alibabaOptions.enableThinking != null
+        ? { enable_thinking: alibabaOptions.enableThinking }
+        : {}),
+      ...(alibabaOptions.thinkingBudget != null
+        ? { thinking_budget: alibabaOptions.thinkingBudget }
+        : {}),
+    };
+  }
+
+  if (!isCustomReasoning(reasoning)) {
+    return {};
+  }
+
+  if (reasoning === 'none') {
+    return { enable_thinking: false };
+  }
+
+  const thinkingBudget = mapReasoningToProviderBudget({
+    reasoning,
+    maxOutputTokens: 16384,
+    maxReasoningBudget: 16384,
+    warnings,
+  });
+
+  return {
+    enable_thinking: true,
+    ...(thinkingBudget != null ? { thinking_budget: thinkingBudget } : {}),
+  };
+}
+
 /**
  * Reference for schemas below:
  * https://www.alibabacloud.com/help/en/model-studio/qwen-api-via-openai-chat-completions

package/src/alibaba-config.ts

@@ -3,7 +3,7 @@ import type { FetchFunction } from '@ai-sdk/provider-utils';
 export interface AlibabaConfig {
   provider: string;
   baseURL: string;
-  headers: () => Record<string, string | undefined>;
+  headers?: () => Record<string, string | undefined>;
   fetch?: FetchFunction;
   includeUsage?: boolean;
 }

package/src/alibaba-error.ts

@@ -0,0 +1,17 @@
+import { createJsonErrorResponseHandler } from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
+
+const alibabaErrorDataSchema = z.object({
+  error: z.object({
+    message: z.string(),
+    code: z.string().nullish(),
+    type: z.string().nullish(),
+  }),
+});
+
+export type AlibabaErrorData = z.infer<typeof alibabaErrorDataSchema>;
+
+export const alibabaFailedResponseHandler = createJsonErrorResponseHandler({
+  errorSchema: alibabaErrorDataSchema,
+  errorToMessage: data => data.error.message,
+});

package/src/alibaba-provider.ts

@@ -1,37 +1,23 @@
 import {
+  NoSuchModelError,
   type Experimental_VideoModelV4,
   type LanguageModelV4,
-  NoSuchModelError,
   type ProviderV4,
 } from '@ai-sdk/provider';
 import {
-  createJsonErrorResponseHandler,
-  type FetchFunction,
   loadApiKey,
   withoutTrailingSlash,
   withUserAgentSuffix,
+  type FetchFunction,
 } from '@ai-sdk/provider-utils';
-import { z } from 'zod/v4';
-import { AlibabaLanguageModel } from './alibaba-chat-language-model';
-import type { AlibabaChatModelId } from './alibaba-chat-options';
+import { AlibabaChatLanguageModel } from './alibaba-chat-language-model';
+import type { AlibabaChatModelId } from './alibaba-chat-language-model-options';
 import { AlibabaVideoModel } from './alibaba-video-model';
 import type { AlibabaVideoModelId } from './alibaba-video-settings';
 import { VERSION } from './version';
 
-export type AlibabaErrorData = z.infer<typeof alibabaErrorDataSchema>;
-
-const alibabaErrorDataSchema = z.object({
-  error: z.object({
-    message: z.string(),
-    code: z.string().nullish(),
-    type: z.string().nullish(),
-  }),
-});
-
-export const alibabaFailedResponseHandler = createJsonErrorResponseHandler({
-  errorSchema: alibabaErrorDataSchema,
-  errorToMessage: data => data.error.message,
-});
+export type { AlibabaErrorData } from './alibaba-error';
+export { alibabaFailedResponseHandler } from './alibaba-error';
 
 export interface AlibabaProvider extends ProviderV4 {
   (modelId: AlibabaChatModelId): LanguageModelV4;
@@ -125,7 +111,7 @@ export function createAlibaba(
     );
 
   const createLanguageModel = (modelId: AlibabaChatModelId) =>
-    new AlibabaLanguageModel(modelId, {
+    new AlibabaChatLanguageModel(modelId, {
       provider: 'alibaba.chat',
       baseURL,
       headers: getHeaders,

package/src/alibaba-video-model-options.ts

@@ -0,0 +1,46 @@
+import { lazySchema, zodSchema } from '@ai-sdk/provider-utils';
+import { z } from 'zod/v4';
+
+export type AlibabaVideoModelOptions = {
+  /** Negative prompt to specify what to avoid (max 500 chars). */
+  negativePrompt?: string | null;
+  /** URL to audio file for audio-video sync (WAV/MP3, 3-30s, max 15MB). */
+  audioUrl?: string | null;
+  /** Enable prompt extension/rewriting for better generation. Defaults to true. */
+  promptExtend?: boolean | null;
+  /** Shot type: 'single' for single-shot or 'multi' for multi-shot narrative. */
+  shotType?: 'single' | 'multi' | null;
+  /** Whether to add watermark to generated video. Defaults to false. */
+  watermark?: boolean | null;
+  /** Enable audio generation (for I2V/R2V models). */
+  audio?: boolean | null;
+  /**
+   * Reference URLs for reference-to-video mode.
+   * Array of URLs to images (0-5) and/or videos (0-3), max 5 total.
+   * Use character identifiers (character1, character2) in prompts to reference them.
+   */
+  referenceUrls?: string[] | null;
+  /** Polling interval in milliseconds. Defaults to 5000 (5 seconds). */
+  pollIntervalMs?: number | null;
+  /** Maximum wait time in milliseconds for video generation. Defaults to 600000 (10 minutes). */
+  pollTimeoutMs?: number | null;
+  [key: string]: unknown;
+};
+
+export const alibabaVideoModelOptionsSchema = lazySchema(() =>
+  zodSchema(
+    z
+      .object({
+        negativePrompt: z.string().nullish(),
+        audioUrl: z.string().nullish(),
+        promptExtend: z.boolean().nullish(),
+        shotType: z.enum(['single', 'multi']).nullish(),
+        watermark: z.boolean().nullish(),
+        audio: z.boolean().nullish(),
+        referenceUrls: z.array(z.string()).nullish(),
+        pollIntervalMs: z.number().positive().nullish(),
+        pollTimeoutMs: z.number().positive().nullish(),
+      })
+      .passthrough(),
+  ),
+);

package/src/alibaba-video-model.ts

@@ -9,62 +9,20 @@ import {
   createJsonErrorResponseHandler,
   createJsonResponseHandler,
   delay,
-  type FetchFunction,
   getFromApi,
-  lazySchema,
   parseProviderOptions,
   postJsonToApi,
-  type Resolvable,
   resolve,
-  zodSchema,
+  type FetchFunction,
+  type Resolvable,
 } from '@ai-sdk/provider-utils';
 import { z } from 'zod/v4';
+import {
+  alibabaVideoModelOptionsSchema,
+  type AlibabaVideoModelOptions,
+} from './alibaba-video-model-options';
 import type { AlibabaVideoModelId } from './alibaba-video-settings';
 
-export type AlibabaVideoModelOptions = {
-  /** Negative prompt to specify what to avoid (max 500 chars). */
-  negativePrompt?: string | null;
-  /** URL to audio file for audio-video sync (WAV/MP3, 3-30s, max 15MB). */
-  audioUrl?: string | null;
-  /** Enable prompt extension/rewriting for better generation. Defaults to true. */
-  promptExtend?: boolean | null;
-  /** Shot type: 'single' for single-shot or 'multi' for multi-shot narrative. */
-  shotType?: 'single' | 'multi' | null;
-  /** Whether to add watermark to generated video. Defaults to false. */
-  watermark?: boolean | null;
-  /** Enable audio generation (for I2V/R2V models). */
-  audio?: boolean | null;
-  /**
-   * Reference URLs for reference-to-video mode.
-   * Array of URLs to images (0-5) and/or videos (0-3), max 5 total.
-   * Use character identifiers (character1, character2) in prompts to reference them.
-   */
-  referenceUrls?: string[] | null;
-  /** Polling interval in milliseconds. Defaults to 5000 (5 seconds). */
-  pollIntervalMs?: number | null;
-  /** Maximum wait time in milliseconds for video generation. Defaults to 600000 (10 minutes). */
-  pollTimeoutMs?: number | null;
-  [key: string]: unknown;
-};
-
-const alibabaVideoModelOptionsSchema = lazySchema(() =>
-  zodSchema(
-    z
-      .object({
-        negativePrompt: z.string().nullish(),
-        audioUrl: z.string().nullish(),
-        promptExtend: z.boolean().nullish(),
-        shotType: z.enum(['single', 'multi']).nullish(),
-        watermark: z.boolean().nullish(),
-        audio: z.boolean().nullish(),
-        referenceUrls: z.array(z.string()).nullish(),
-        pollIntervalMs: z.number().positive().nullish(),
-        pollTimeoutMs: z.number().positive().nullish(),
-      })
-      .passthrough(),
-  ),
-);
-
 interface AlibabaVideoModelConfig {
   provider: string;
   baseURL: string;

package/src/convert-to-alibaba-chat-messages.ts

@@ -1,22 +1,28 @@
 import {
-  type LanguageModelV4DataContent,
-  type LanguageModelV4Prompt,
   UnsupportedFunctionalityError,
+  type LanguageModelV4FilePart,
+  type LanguageModelV4Prompt,
 } from '@ai-sdk/provider';
-import { convertToBase64 } from '@ai-sdk/provider-utils';
+import {
+  convertToBase64,
+  getTopLevelMediaType,
+  resolveFullMediaType,
+} from '@ai-sdk/provider-utils';
 import type { AlibabaChatPrompt } from './alibaba-chat-prompt';
 import type { CacheControlValidator } from './get-cache-control';
 
-function formatImageUrl({
-  data,
-  mediaType,
-}: {
-  data: LanguageModelV4DataContent;
-  mediaType: string;
-}): string {
-  return data instanceof URL
-    ? data.toString()
-    : `data:${mediaType};base64,${convertToBase64(data as Uint8Array)}`;
+function formatImageUrl({ part }: { part: LanguageModelV4FilePart }): string {
+  if (part.data.type === 'url') {
+    return part.data.url.toString();
+  }
+
+  if (part.data.type === 'data') {
+    return `data:${resolveFullMediaType({ part })};base64,${convertToBase64(part.data.data)}`;
+  }
+
+  throw new UnsupportedFunctionalityError({
+    functionality: `file part data type ${part.data.type}`,
+  });
 }
 
 export function convertToAlibabaChatMessages({
@@ -73,25 +79,35 @@ export function convertToAlibabaChatMessages({
               }
 
               case 'file': {
-                if (part.mediaType.startsWith('image/')) {
-                  const mediaType =
-                    part.mediaType === 'image/*'
-                      ? 'image/jpeg'
-                      : part.mediaType;
-
-                  return {
-                    type: 'image_url',
-                    image_url: {
-                      url: formatImageUrl({ data: part.data, mediaType }),
-                    },
-                    ...(partCacheControl
-                      ? { cache_control: partCacheControl }
-                      : {}),
-                  };
-                } else {
-                  throw new UnsupportedFunctionalityError({
-                    functionality: 'Only image file parts are supported',
-                  });
+                switch (part.data.type) {
+                  case 'reference': {
+                    throw new UnsupportedFunctionalityError({
+                      functionality: 'file parts with provider references',
+                    });
+                  }
+                  case 'text': {
+                    throw new UnsupportedFunctionalityError({
+                      functionality: 'text file parts',
+                    });
+                  }
+                  case 'url':
+                  case 'data': {
+                    if (getTopLevelMediaType(part.mediaType) === 'image') {
+                      return {
+                        type: 'image_url',
+                        image_url: {
+                          url: formatImageUrl({ part }),
+                        },
+                        ...(partCacheControl
+                          ? { cache_control: partCacheControl }
+                          : {}),
+                      };
+                    } else {
+                      throw new UnsupportedFunctionalityError({
+                        functionality: 'Only image file parts are supported',
+                      });
+                    }
+                  }
                 }
               }
             }
@@ -167,7 +183,7 @@ export function convertToAlibabaChatMessages({
               contentValue = output.value;
               break;
             case 'execution-denied':
-              contentValue = output.reason ?? 'Tool execution denied.';
+              contentValue = output.reason ?? 'Tool call execution denied.';
               break;
             case 'content':
             case 'json':

package/src/index.ts

@@ -1,9 +1,11 @@
 export type {
   AlibabaChatModelId,
-  AlibabaLanguageModelOptions,
-  /** @deprecated Use `AlibabaLanguageModelOptions` instead. */
-  AlibabaLanguageModelOptions as AlibabaProviderOptions,
-} from './alibaba-chat-options';
+  AlibabaLanguageModelChatOptions,
+  /** @deprecated Use `AlibabaLanguageModelChatOptions` instead. */
+  AlibabaLanguageModelChatOptions as AlibabaLanguageModelOptions,
+  /** @deprecated Use `AlibabaLanguageModelChatOptions` instead. */
+  AlibabaLanguageModelChatOptions as AlibabaProviderOptions,
+} from './alibaba-chat-language-model-options';
 export type { AlibabaCacheControl } from './alibaba-chat-prompt';
 export {
   type AlibabaProvider,
@@ -15,7 +17,7 @@ export type {
   AlibabaVideoModelOptions,
   /** @deprecated Use `AlibabaVideoModelOptions` instead. */
   AlibabaVideoModelOptions as AlibabaVideoProviderOptions,
-} from './alibaba-video-model';
+} from './alibaba-video-model-options';
 export type { AlibabaVideoModelId } from './alibaba-video-settings';
 export type { AlibabaUsage } from './convert-alibaba-usage';
 export { VERSION } from './version';