@ai-sdk/cerebras 3.0.32 → 3.0.34

package/CHANGELOG.md

@@ -1,5 +1,20 @@
 # @ai-sdk/cerebras
 
+## 3.0.34
+
+### Patch Changes
+
+- d6a521a: feat(cerebras): add typed provider options and send `maxOutputTokens` as `max_completion_tokens` field
+- Updated dependencies [99989ba]
+  - @ai-sdk/openai-compatible@3.0.34
+
+## 3.0.33
+
+### Patch Changes
+
+- Updated dependencies [d68139c]
+  - @ai-sdk/openai-compatible@3.0.33
+
 ## 3.0.32
 
 ### Patch Changes

package/dist/index.d.ts

@@ -51,6 +51,43 @@ interface CerebrasProvider extends ProviderV4 {
 declare function createCerebras(options?: CerebrasProviderSettings): CerebrasProvider;
 declare const cerebras: CerebrasProvider;
 
+declare const cerebrasLanguageModelChatOptions: z.ZodObject<{
+    user: z.ZodOptional<z.ZodString>;
+    strictJsonSchema: z.ZodOptional<z.ZodBoolean>;
+    parallelToolCalls: z.ZodOptional<z.ZodBoolean>;
+    logprobs: z.ZodOptional<z.ZodBoolean>;
+    topLogprobs: z.ZodOptional<z.ZodNumber>;
+    logitBias: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodNumber>>;
+    serviceTier: z.ZodOptional<z.ZodEnum<{
+        default: "default";
+        auto: "auto";
+        flex: "flex";
+        priority: "priority";
+    }>>;
+    reasoningEffort: z.ZodOptional<z.ZodEnum<{
+        none: "none";
+        low: "low";
+        medium: "medium";
+        high: "high";
+    }>>;
+    reasoningFormat: z.ZodOptional<z.ZodEnum<{
+        none: "none";
+        parsed: "parsed";
+        text_parsed: "text_parsed";
+        raw: "raw";
+        hidden: "hidden";
+    }>>;
+    prediction: z.ZodOptional<z.ZodObject<{
+        type: z.ZodLiteral<"content">;
+        content: z.ZodUnion<readonly [z.ZodString, z.ZodArray<z.ZodObject<{
+            type: z.ZodLiteral<"text">;
+            text: z.ZodString;
+        }, z.core.$strip>>]>;
+    }, z.core.$strip>>;
+    promptCacheKey: z.ZodOptional<z.ZodString>;
+}, z.core.$strip>;
+type CerebrasLanguageModelChatOptions = z.infer<typeof cerebrasLanguageModelChatOptions>;
+
 declare const VERSION: string;
 
-export { type CerebrasErrorData, type CerebrasProvider, type CerebrasProviderSettings, VERSION, cerebras, createCerebras };
+export { type CerebrasErrorData, type CerebrasLanguageModelChatOptions, type CerebrasProvider, type CerebrasProviderSettings, VERSION, cerebras, createCerebras };

package/dist/index.js

@@ -96,7 +96,7 @@ var CerebrasChatLanguageModel = class _CerebrasChatLanguageModel extends OpenAIC
 import { z } from "zod/v4";
 
 // src/version.ts
-var VERSION = true ? "3.0.32" : "0.0.0-test";
+var VERSION = true ? "3.0.34" : "0.0.0-test";
 
 // src/cerebras-provider.ts
 var cerebrasErrorSchema = z.object({
@@ -110,8 +110,33 @@ var cerebrasErrorStructure = {
   errorToMessage: (data) => data.message
 };
 function transformCerebrasRequestBody(args) {
+  const {
+    max_tokens: maxTokens,
+    parallelToolCalls,
+    topLogprobs,
+    logitBias,
+    serviceTier,
+    reasoningFormat,
+    promptCacheKey,
+    ...restArgs
+  } = args;
   return {
-    ...args,
+    ...restArgs,
+    ...maxTokens !== void 0 && {
+      max_completion_tokens: maxTokens
+    },
+    ...parallelToolCalls !== void 0 && {
+      parallel_tool_calls: parallelToolCalls
+    },
+    ...topLogprobs !== void 0 && { top_logprobs: topLogprobs },
+    ...logitBias !== void 0 && { logit_bias: logitBias },
+    ...serviceTier !== void 0 && { service_tier: serviceTier },
+    ...reasoningFormat !== void 0 && {
+      reasoning_format: reasoningFormat
+    },
+    ...promptCacheKey !== void 0 && {
+      prompt_cache_key: promptCacheKey
+    },
     messages: Array.isArray(args.messages) ? args.messages.map((message) => {
       if (message == null || typeof message !== "object" || message.role !== "assistant" || !("reasoning_content" in message)) {
         return message;

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../src/cerebras-provider.ts","../src/cerebras-chat-language-model.ts","../src/version.ts"],"sourcesContent":["import type { ProviderErrorStructure } from '@ai-sdk/openai-compatible';\nimport {\n  NoSuchModelError,\n  type LanguageModelV4,\n  type ProviderV4,\n} from '@ai-sdk/provider';\nimport {\n  loadApiKey,\n  withoutTrailingSlash,\n  withUserAgentSuffix,\n  type FetchFunction,\n} from '@ai-sdk/provider-utils';\nimport { CerebrasChatLanguageModel } from './cerebras-chat-language-model';\nimport type { CerebrasChatModelId } from './cerebras-chat-options';\nimport { z } from 'zod/v4';\nimport { VERSION } from './version';\n\n// Add error schema and structure\nconst cerebrasErrorSchema = z.object({\n  message: z.string(),\n  type: z.string(),\n  param: z.string(),\n  code: z.string(),\n});\n\nexport type CerebrasErrorData = z.infer<typeof cerebrasErrorSchema>;\n\nconst cerebrasErrorStructure: ProviderErrorStructure<CerebrasErrorData> = {\n  errorSchema: cerebrasErrorSchema,\n  errorToMessage: data => data.message,\n};\n\n/**\n * Cerebras expects assistant reasoning history in the `reasoning` field, while\n * the shared OpenAI-compatible converter serializes it as `reasoning_content`.\n */\nfunction transformCerebrasRequestBody(\n  args: Record<string, any>,\n): Record<string, any> {\n  return {\n    ...args,\n    messages: Array.isArray(args.messages)\n      ? args.messages.map(message => {\n          if (\n            message == null ||\n            typeof message !== 'object' ||\n            message.role !== 'assistant' ||\n            !('reasoning_content' in message)\n          ) {\n            return message;\n          }\n\n          const { reasoning_content, ...rest } = message;\n\n          return {\n            ...rest,\n            ...(!('reasoning' in rest) && reasoning_content !== undefined\n              ? { reasoning: reasoning_content }\n              : {}),\n          };\n        })\n      : args.messages,\n  };\n}\n\nexport interface CerebrasProviderSettings {\n  /**\n   * Cerebras API key.\n   */\n  apiKey?: string;\n  /**\n   * Base URL for the API calls.\n   */\n  baseURL?: string;\n  /**\n   * Custom headers to include in the requests.\n   */\n  headers?: Record<string, string>;\n  /**\n   * Custom fetch implementation. You can use it as a middleware to intercept requests,\n   * or to provide a custom fetch implementation for e.g. testing.\n   */\n  fetch?: FetchFunction;\n}\n\nexport interface CerebrasProvider extends ProviderV4 {\n  /**\n   * Creates a Cerebras model for text generation.\n   */\n  (modelId: CerebrasChatModelId): LanguageModelV4;\n\n  /**\n   * Creates a Cerebras model for text generation.\n   */\n  languageModel(modelId: CerebrasChatModelId): LanguageModelV4;\n\n  /**\n   * Creates a Cerebras chat model for text generation.\n   */\n  chat(modelId: CerebrasChatModelId): LanguageModelV4;\n\n  /**\n   * @deprecated Use `embeddingModel` instead.\n   */\n  textEmbeddingModel(modelId: string): never;\n}\n\nexport function createCerebras(\n  options: CerebrasProviderSettings = {},\n): CerebrasProvider {\n  const baseURL = withoutTrailingSlash(\n    options.baseURL ?? 'https://api.cerebras.ai/v1',\n  );\n  const getHeaders = () =>\n    withUserAgentSuffix(\n      {\n        Authorization: `Bearer ${loadApiKey({\n          apiKey: options.apiKey,\n          environmentVariableName: 'CEREBRAS_API_KEY',\n          description: 'Cerebras API key',\n        })}`,\n        ...options.headers,\n      },\n      `ai-sdk/cerebras/${VERSION}`,\n    );\n\n  const createLanguageModel = (modelId: CerebrasChatModelId) => {\n    return new CerebrasChatLanguageModel(modelId, {\n      provider: `cerebras.chat`,\n      url: ({ path }) => `${baseURL}${path}`,\n      headers: getHeaders,\n      fetch: options.fetch,\n      errorStructure: cerebrasErrorStructure,\n      supportsStructuredOutputs: true,\n      transformRequestBody: transformCerebrasRequestBody,\n    });\n  };\n\n  const provider = (modelId: CerebrasChatModelId) =>\n    createLanguageModel(modelId);\n\n  provider.specificationVersion = 'v4' as const;\n  provider.languageModel = createLanguageModel;\n  provider.chat = createLanguageModel;\n\n  provider.embeddingModel = (modelId: string) => {\n    throw new NoSuchModelError({ modelId, modelType: 'embeddingModel' });\n  };\n  provider.textEmbeddingModel = provider.embeddingModel;\n  provider.imageModel = (modelId: string) => {\n    throw new NoSuchModelError({ modelId, modelType: 'imageModel' });\n  };\n\n  return provider;\n}\n\nexport const cerebras = createCerebras();\n","import { OpenAICompatibleChatLanguageModel } from '@ai-sdk/openai-compatible';\nimport type {\n  LanguageModelV4,\n  LanguageModelV4CallOptions,\n  LanguageModelV4GenerateResult,\n  LanguageModelV4StreamPart,\n  LanguageModelV4StreamResult,\n} from '@ai-sdk/provider';\nimport {\n  serializeModelOptions,\n  WORKFLOW_DESERIALIZE,\n  WORKFLOW_SERIALIZE,\n} from '@ai-sdk/provider-utils';\nimport type { CerebrasChatModelId } from './cerebras-chat-language-model-options';\n\ntype CerebrasChatConfig = ConstructorParameters<\n  typeof OpenAICompatibleChatLanguageModel\n>[1];\n\nfunction isStructuredOutputWithToolCallsFinishReason({\n  rawFinishReason,\n  hasText,\n  responseFormatType,\n}: {\n  rawFinishReason: string | undefined;\n  hasText: boolean;\n  responseFormatType:\n    | NonNullable<LanguageModelV4CallOptions['responseFormat']>['type']\n    | undefined;\n}) {\n  return (\n    responseFormatType === 'json' && rawFinishReason === 'tool_calls' && hasText\n  );\n}\n\nexport class CerebrasChatLanguageModel\n  extends OpenAICompatibleChatLanguageModel\n  implements LanguageModelV4\n{\n  static [WORKFLOW_SERIALIZE](model: CerebrasChatLanguageModel) {\n    return serializeModelOptions({\n      modelId: model.modelId,\n      config: model.config,\n    });\n  }\n\n  static [WORKFLOW_DESERIALIZE](options: {\n    modelId: CerebrasChatModelId;\n    config: CerebrasChatConfig;\n  }) {\n    return new CerebrasChatLanguageModel(options.modelId, options.config);\n  }\n\n  async doGenerate(\n    options: LanguageModelV4CallOptions,\n  ): Promise<LanguageModelV4GenerateResult> {\n    const result = await super.doGenerate(options);\n\n    if (\n      !isStructuredOutputWithToolCallsFinishReason({\n        rawFinishReason: result.finishReason.raw,\n        hasText: result.content.some(\n          part => part.type === 'text' && part.text.length > 0,\n        ),\n        responseFormatType: options.responseFormat?.type,\n      })\n    ) {\n      return result;\n    }\n\n    return {\n      ...result,\n      // Cerebras GLM can return valid structured output text while also\n      // repeating a tool call. Treat that mixed response as the final answer.\n      content: result.content.filter(part => part.type !== 'tool-call'),\n      finishReason: {\n        ...result.finishReason,\n        unified: 'stop',\n      },\n    };\n  }\n\n  async doStream(\n    options: LanguageModelV4CallOptions,\n  ): Promise<LanguageModelV4StreamResult> {\n    const result = await super.doStream(options);\n    let hasText = false;\n\n    return {\n      ...result,\n      stream: result.stream.pipeThrough(\n        new TransformStream<\n          LanguageModelV4StreamPart,\n          LanguageModelV4StreamPart\n        >({\n          transform(part, controller) {\n            if (part.type === 'text-delta' && part.delta.length > 0) {\n              hasText = true;\n            }\n\n            if (\n              part.type === 'finish' &&\n              isStructuredOutputWithToolCallsFinishReason({\n                rawFinishReason: part.finishReason.raw,\n                hasText,\n                responseFormatType: options.responseFormat?.type,\n              })\n            ) {\n              controller.enqueue({\n                ...part,\n                finishReason: {\n                  ...part.finishReason,\n                  unified: 'stop',\n                },\n              });\n              return;\n            }\n\n            if (\n              options.responseFormat?.type === 'json' &&\n              hasText &&\n              (part.type === 'tool-input-start' ||\n                part.type === 'tool-input-delta' ||\n                part.type === 'tool-input-end' ||\n                part.type === 'tool-call')\n            ) {\n              return;\n            }\n\n            controller.enqueue(part);\n          },\n        }),\n      ),\n    };\n  }\n}\n","// Version string of this package injected at build time.\ndeclare const __PACKAGE_VERSION__: string | undefined;\nexport const VERSION: string =\n  typeof __PACKAGE_VERSION__ !== 'undefined'\n    ? __PACKAGE_VERSION__\n    : '0.0.0-test';\n"],"mappings":";AACA;AAAA,EACE;AAAA,OAGK;AACP;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,OAEK;;;ACXP,SAAS,yCAAyC;AAQlD;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAOP,SAAS,4CAA4C;AAAA,EACnD;AAAA,EACA;AAAA,EACA;AACF,GAMG;AACD,SACE,uBAAuB,UAAU,oBAAoB,gBAAgB;AAEzE;AAEO,IAAM,4BAAN,MAAM,mCACH,kCAEV;AAAA,EACE,QAAQ,kBAAkB,EAAE,OAAkC;AAC5D,WAAO,sBAAsB;AAAA,MAC3B,SAAS,MAAM;AAAA,MACf,QAAQ,MAAM;AAAA,IAChB,CAAC;AAAA,EACH;AAAA,EAEA,QAAQ,oBAAoB,EAAE,SAG3B;AACD,WAAO,IAAI,2BAA0B,QAAQ,SAAS,QAAQ,MAAM;AAAA,EACtE;AAAA,EAEA,MAAM,WACJ,SACwC;AAvD5C;AAwDI,UAAM,SAAS,MAAM,MAAM,WAAW,OAAO;AAE7C,QACE,CAAC,4CAA4C;AAAA,MAC3C,iBAAiB,OAAO,aAAa;AAAA,MACrC,SAAS,OAAO,QAAQ;AAAA,QACtB,UAAQ,KAAK,SAAS,UAAU,KAAK,KAAK,SAAS;AAAA,MACrD;AAAA,MACA,qBAAoB,aAAQ,mBAAR,mBAAwB;AAAA,IAC9C,CAAC,GACD;AACA,aAAO;AAAA,IACT;AAEA,WAAO;AAAA,MACL,GAAG;AAAA;AAAA;AAAA,MAGH,SAAS,OAAO,QAAQ,OAAO,UAAQ,KAAK,SAAS,WAAW;AAAA,MAChE,cAAc;AAAA,QACZ,GAAG,OAAO;AAAA,QACV,SAAS;AAAA,MACX;AAAA,IACF;AAAA,EACF;AAAA,EAEA,MAAM,SACJ,SACsC;AACtC,UAAM,SAAS,MAAM,MAAM,SAAS,OAAO;AAC3C,QAAI,UAAU;AAEd,WAAO;AAAA,MACL,GAAG;AAAA,MACH,QAAQ,OAAO,OAAO;AAAA,QACpB,IAAI,gBAGF;AAAA,UACA,UAAU,MAAM,YAAY;AA/FtC;AAgGY,gBAAI,KAAK,SAAS,gBAAgB,KAAK,MAAM,SAAS,GAAG;AACvD,wBAAU;AAAA,YACZ;AAEA,gBACE,KAAK,SAAS,YACd,4CAA4C;AAAA,cAC1C,iBAAiB,KAAK,aAAa;AAAA,cACnC;AAAA,cACA,qBAAoB,aAAQ,mBAAR,mBAAwB;AAAA,YAC9C,CAAC,GACD;AACA,yBAAW,QAAQ;AAAA,gBACjB,GAAG;AAAA,gBACH,cAAc;AAAA,kBACZ,GAAG,KAAK;AAAA,kBACR,SAAS;AAAA,gBACX;AAAA,cACF,CAAC;AACD;AAAA,YACF;AAEA,kBACE,aAAQ,mBAAR,mBAAwB,UAAS,UACjC,YACC,KAAK,SAAS,sBACb,KAAK,SAAS,sBACd,KAAK,SAAS,oBACd,KAAK,SAAS,cAChB;AACA;AAAA,YACF;AAEA,uBAAW,QAAQ,IAAI;AAAA,UACzB;AAAA,QACF,CAAC;AAAA,MACH;AAAA,IACF;AAAA,EACF;AACF;;;ADzHA,SAAS,SAAS;;;AEZX,IAAM,UACX,OACI,WACA;;;AFaN,IAAM,sBAAsB,EAAE,OAAO;AAAA,EACnC,SAAS,EAAE,OAAO;AAAA,EAClB,MAAM,EAAE,OAAO;AAAA,EACf,OAAO,EAAE,OAAO;AAAA,EAChB,MAAM,EAAE,OAAO;AACjB,CAAC;AAID,IAAM,yBAAoE;AAAA,EACxE,aAAa;AAAA,EACb,gBAAgB,UAAQ,KAAK;AAC/B;AAMA,SAAS,6BACP,MACqB;AACrB,SAAO;AAAA,IACL,GAAG;AAAA,IACH,UAAU,MAAM,QAAQ,KAAK,QAAQ,IACjC,KAAK,SAAS,IAAI,aAAW;AAC3B,UACE,WAAW,QACX,OAAO,YAAY,YACnB,QAAQ,SAAS,eACjB,EAAE,uBAAuB,UACzB;AACA,eAAO;AAAA,MACT;AAEA,YAAM,EAAE,mBAAmB,GAAG,KAAK,IAAI;AAEvC,aAAO;AAAA,QACL,GAAG;AAAA,QACH,GAAI,EAAE,eAAe,SAAS,sBAAsB,SAChD,EAAE,WAAW,kBAAkB,IAC/B,CAAC;AAAA,MACP;AAAA,IACF,CAAC,IACD,KAAK;AAAA,EACX;AACF;AA4CO,SAAS,eACd,UAAoC,CAAC,GACnB;AA7GpB;AA8GE,QAAM,UAAU;AAAA,KACd,aAAQ,YAAR,YAAmB;AAAA,EACrB;AACA,QAAM,aAAa,MACjB;AAAA,IACE;AAAA,MACE,eAAe,UAAU,WAAW;AAAA,QAClC,QAAQ,QAAQ;AAAA,QAChB,yBAAyB;AAAA,QACzB,aAAa;AAAA,MACf,CAAC,CAAC;AAAA,MACF,GAAG,QAAQ;AAAA,IACb;AAAA,IACA,mBAAmB,OAAO;AAAA,EAC5B;AAEF,QAAM,sBAAsB,CAAC,YAAiC;AAC5D,WAAO,IAAI,0BAA0B,SAAS;AAAA,MAC5C,UAAU;AAAA,MACV,KAAK,CAAC,EAAE,KAAK,MAAM,GAAG,OAAO,GAAG,IAAI;AAAA,MACpC,SAAS;AAAA,MACT,OAAO,QAAQ;AAAA,MACf,gBAAgB;AAAA,MAChB,2BAA2B;AAAA,MAC3B,sBAAsB;AAAA,IACxB,CAAC;AAAA,EACH;AAEA,QAAM,WAAW,CAAC,YAChB,oBAAoB,OAAO;AAE7B,WAAS,uBAAuB;AAChC,WAAS,gBAAgB;AACzB,WAAS,OAAO;AAEhB,WAAS,iBAAiB,CAAC,YAAoB;AAC7C,UAAM,IAAI,iBAAiB,EAAE,SAAS,WAAW,iBAAiB,CAAC;AAAA,EACrE;AACA,WAAS,qBAAqB,SAAS;AACvC,WAAS,aAAa,CAAC,YAAoB;AACzC,UAAM,IAAI,iBAAiB,EAAE,SAAS,WAAW,aAAa,CAAC;AAAA,EACjE;AAEA,SAAO;AACT;AAEO,IAAM,WAAW,eAAe;","names":[]}
+{"version":3,"sources":["../src/cerebras-provider.ts","../src/cerebras-chat-language-model.ts","../src/version.ts"],"sourcesContent":["import type { ProviderErrorStructure } from '@ai-sdk/openai-compatible';\nimport {\n  NoSuchModelError,\n  type LanguageModelV4,\n  type ProviderV4,\n} from '@ai-sdk/provider';\nimport {\n  loadApiKey,\n  withoutTrailingSlash,\n  withUserAgentSuffix,\n  type FetchFunction,\n} from '@ai-sdk/provider-utils';\nimport { CerebrasChatLanguageModel } from './cerebras-chat-language-model';\nimport type { CerebrasChatModelId } from './cerebras-chat-options';\nimport { z } from 'zod/v4';\nimport { VERSION } from './version';\n\n// Add error schema and structure\nconst cerebrasErrorSchema = z.object({\n  message: z.string(),\n  type: z.string(),\n  param: z.string(),\n  code: z.string(),\n});\n\nexport type CerebrasErrorData = z.infer<typeof cerebrasErrorSchema>;\n\nconst cerebrasErrorStructure: ProviderErrorStructure<CerebrasErrorData> = {\n  errorSchema: cerebrasErrorSchema,\n  errorToMessage: data => data.message,\n};\n\n/**\n * Cerebras expects assistant reasoning history in the `reasoning` field, while\n * the shared OpenAI-compatible converter serializes it as `reasoning_content`.\n */\nfunction transformCerebrasRequestBody(\n  args: Record<string, any>,\n): Record<string, any> {\n  const {\n    max_tokens: maxTokens,\n    parallelToolCalls,\n    topLogprobs,\n    logitBias,\n    serviceTier,\n    reasoningFormat,\n    promptCacheKey,\n    ...restArgs\n  } = args;\n\n  return {\n    ...restArgs,\n    ...(maxTokens !== undefined && {\n      max_completion_tokens: maxTokens,\n    }),\n    ...(parallelToolCalls !== undefined && {\n      parallel_tool_calls: parallelToolCalls,\n    }),\n    ...(topLogprobs !== undefined && { top_logprobs: topLogprobs }),\n    ...(logitBias !== undefined && { logit_bias: logitBias }),\n    ...(serviceTier !== undefined && { service_tier: serviceTier }),\n    ...(reasoningFormat !== undefined && {\n      reasoning_format: reasoningFormat,\n    }),\n    ...(promptCacheKey !== undefined && {\n      prompt_cache_key: promptCacheKey,\n    }),\n    messages: Array.isArray(args.messages)\n      ? args.messages.map(message => {\n          if (\n            message == null ||\n            typeof message !== 'object' ||\n            message.role !== 'assistant' ||\n            !('reasoning_content' in message)\n          ) {\n            return message;\n          }\n\n          const { reasoning_content, ...rest } = message;\n\n          return {\n            ...rest,\n            ...(!('reasoning' in rest) && reasoning_content !== undefined\n              ? { reasoning: reasoning_content }\n              : {}),\n          };\n        })\n      : args.messages,\n  };\n}\n\nexport interface CerebrasProviderSettings {\n  /**\n   * Cerebras API key.\n   */\n  apiKey?: string;\n  /**\n   * Base URL for the API calls.\n   */\n  baseURL?: string;\n  /**\n   * Custom headers to include in the requests.\n   */\n  headers?: Record<string, string>;\n  /**\n   * Custom fetch implementation. You can use it as a middleware to intercept requests,\n   * or to provide a custom fetch implementation for e.g. testing.\n   */\n  fetch?: FetchFunction;\n}\n\nexport interface CerebrasProvider extends ProviderV4 {\n  /**\n   * Creates a Cerebras model for text generation.\n   */\n  (modelId: CerebrasChatModelId): LanguageModelV4;\n\n  /**\n   * Creates a Cerebras model for text generation.\n   */\n  languageModel(modelId: CerebrasChatModelId): LanguageModelV4;\n\n  /**\n   * Creates a Cerebras chat model for text generation.\n   */\n  chat(modelId: CerebrasChatModelId): LanguageModelV4;\n\n  /**\n   * @deprecated Use `embeddingModel` instead.\n   */\n  textEmbeddingModel(modelId: string): never;\n}\n\nexport function createCerebras(\n  options: CerebrasProviderSettings = {},\n): CerebrasProvider {\n  const baseURL = withoutTrailingSlash(\n    options.baseURL ?? 'https://api.cerebras.ai/v1',\n  );\n  const getHeaders = () =>\n    withUserAgentSuffix(\n      {\n        Authorization: `Bearer ${loadApiKey({\n          apiKey: options.apiKey,\n          environmentVariableName: 'CEREBRAS_API_KEY',\n          description: 'Cerebras API key',\n        })}`,\n        ...options.headers,\n      },\n      `ai-sdk/cerebras/${VERSION}`,\n    );\n\n  const createLanguageModel = (modelId: CerebrasChatModelId) => {\n    return new CerebrasChatLanguageModel(modelId, {\n      provider: `cerebras.chat`,\n      url: ({ path }) => `${baseURL}${path}`,\n      headers: getHeaders,\n      fetch: options.fetch,\n      errorStructure: cerebrasErrorStructure,\n      supportsStructuredOutputs: true,\n      transformRequestBody: transformCerebrasRequestBody,\n    });\n  };\n\n  const provider = (modelId: CerebrasChatModelId) =>\n    createLanguageModel(modelId);\n\n  provider.specificationVersion = 'v4' as const;\n  provider.languageModel = createLanguageModel;\n  provider.chat = createLanguageModel;\n\n  provider.embeddingModel = (modelId: string) => {\n    throw new NoSuchModelError({ modelId, modelType: 'embeddingModel' });\n  };\n  provider.textEmbeddingModel = provider.embeddingModel;\n  provider.imageModel = (modelId: string) => {\n    throw new NoSuchModelError({ modelId, modelType: 'imageModel' });\n  };\n\n  return provider;\n}\n\nexport const cerebras = createCerebras();\n","import { OpenAICompatibleChatLanguageModel } from '@ai-sdk/openai-compatible';\nimport type {\n  LanguageModelV4,\n  LanguageModelV4CallOptions,\n  LanguageModelV4GenerateResult,\n  LanguageModelV4StreamPart,\n  LanguageModelV4StreamResult,\n} from '@ai-sdk/provider';\nimport {\n  serializeModelOptions,\n  WORKFLOW_DESERIALIZE,\n  WORKFLOW_SERIALIZE,\n} from '@ai-sdk/provider-utils';\nimport type { CerebrasChatModelId } from './cerebras-chat-language-model-options';\n\ntype CerebrasChatConfig = ConstructorParameters<\n  typeof OpenAICompatibleChatLanguageModel\n>[1];\n\nfunction isStructuredOutputWithToolCallsFinishReason({\n  rawFinishReason,\n  hasText,\n  responseFormatType,\n}: {\n  rawFinishReason: string | undefined;\n  hasText: boolean;\n  responseFormatType:\n    | NonNullable<LanguageModelV4CallOptions['responseFormat']>['type']\n    | undefined;\n}) {\n  return (\n    responseFormatType === 'json' && rawFinishReason === 'tool_calls' && hasText\n  );\n}\n\nexport class CerebrasChatLanguageModel\n  extends OpenAICompatibleChatLanguageModel\n  implements LanguageModelV4\n{\n  static [WORKFLOW_SERIALIZE](model: CerebrasChatLanguageModel) {\n    return serializeModelOptions({\n      modelId: model.modelId,\n      config: model.config,\n    });\n  }\n\n  static [WORKFLOW_DESERIALIZE](options: {\n    modelId: CerebrasChatModelId;\n    config: CerebrasChatConfig;\n  }) {\n    return new CerebrasChatLanguageModel(options.modelId, options.config);\n  }\n\n  async doGenerate(\n    options: LanguageModelV4CallOptions,\n  ): Promise<LanguageModelV4GenerateResult> {\n    const result = await super.doGenerate(options);\n\n    if (\n      !isStructuredOutputWithToolCallsFinishReason({\n        rawFinishReason: result.finishReason.raw,\n        hasText: result.content.some(\n          part => part.type === 'text' && part.text.length > 0,\n        ),\n        responseFormatType: options.responseFormat?.type,\n      })\n    ) {\n      return result;\n    }\n\n    return {\n      ...result,\n      // Cerebras GLM can return valid structured output text while also\n      // repeating a tool call. Treat that mixed response as the final answer.\n      content: result.content.filter(part => part.type !== 'tool-call'),\n      finishReason: {\n        ...result.finishReason,\n        unified: 'stop',\n      },\n    };\n  }\n\n  async doStream(\n    options: LanguageModelV4CallOptions,\n  ): Promise<LanguageModelV4StreamResult> {\n    const result = await super.doStream(options);\n    let hasText = false;\n\n    return {\n      ...result,\n      stream: result.stream.pipeThrough(\n        new TransformStream<\n          LanguageModelV4StreamPart,\n          LanguageModelV4StreamPart\n        >({\n          transform(part, controller) {\n            if (part.type === 'text-delta' && part.delta.length > 0) {\n              hasText = true;\n            }\n\n            if (\n              part.type === 'finish' &&\n              isStructuredOutputWithToolCallsFinishReason({\n                rawFinishReason: part.finishReason.raw,\n                hasText,\n                responseFormatType: options.responseFormat?.type,\n              })\n            ) {\n              controller.enqueue({\n                ...part,\n                finishReason: {\n                  ...part.finishReason,\n                  unified: 'stop',\n                },\n              });\n              return;\n            }\n\n            if (\n              options.responseFormat?.type === 'json' &&\n              hasText &&\n              (part.type === 'tool-input-start' ||\n                part.type === 'tool-input-delta' ||\n                part.type === 'tool-input-end' ||\n                part.type === 'tool-call')\n            ) {\n              return;\n            }\n\n            controller.enqueue(part);\n          },\n        }),\n      ),\n    };\n  }\n}\n","// Version string of this package injected at build time.\ndeclare const __PACKAGE_VERSION__: string | undefined;\nexport const VERSION: string =\n  typeof __PACKAGE_VERSION__ !== 'undefined'\n    ? __PACKAGE_VERSION__\n    : '0.0.0-test';\n"],"mappings":";AACA;AAAA,EACE;AAAA,OAGK;AACP;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,OAEK;;;ACXP,SAAS,yCAAyC;AAQlD;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAOP,SAAS,4CAA4C;AAAA,EACnD;AAAA,EACA;AAAA,EACA;AACF,GAMG;AACD,SACE,uBAAuB,UAAU,oBAAoB,gBAAgB;AAEzE;AAEO,IAAM,4BAAN,MAAM,mCACH,kCAEV;AAAA,EACE,QAAQ,kBAAkB,EAAE,OAAkC;AAC5D,WAAO,sBAAsB;AAAA,MAC3B,SAAS,MAAM;AAAA,MACf,QAAQ,MAAM;AAAA,IAChB,CAAC;AAAA,EACH;AAAA,EAEA,QAAQ,oBAAoB,EAAE,SAG3B;AACD,WAAO,IAAI,2BAA0B,QAAQ,SAAS,QAAQ,MAAM;AAAA,EACtE;AAAA,EAEA,MAAM,WACJ,SACwC;AAvD5C;AAwDI,UAAM,SAAS,MAAM,MAAM,WAAW,OAAO;AAE7C,QACE,CAAC,4CAA4C;AAAA,MAC3C,iBAAiB,OAAO,aAAa;AAAA,MACrC,SAAS,OAAO,QAAQ;AAAA,QACtB,UAAQ,KAAK,SAAS,UAAU,KAAK,KAAK,SAAS;AAAA,MACrD;AAAA,MACA,qBAAoB,aAAQ,mBAAR,mBAAwB;AAAA,IAC9C,CAAC,GACD;AACA,aAAO;AAAA,IACT;AAEA,WAAO;AAAA,MACL,GAAG;AAAA;AAAA;AAAA,MAGH,SAAS,OAAO,QAAQ,OAAO,UAAQ,KAAK,SAAS,WAAW;AAAA,MAChE,cAAc;AAAA,QACZ,GAAG,OAAO;AAAA,QACV,SAAS;AAAA,MACX;AAAA,IACF;AAAA,EACF;AAAA,EAEA,MAAM,SACJ,SACsC;AACtC,UAAM,SAAS,MAAM,MAAM,SAAS,OAAO;AAC3C,QAAI,UAAU;AAEd,WAAO;AAAA,MACL,GAAG;AAAA,MACH,QAAQ,OAAO,OAAO;AAAA,QACpB,IAAI,gBAGF;AAAA,UACA,UAAU,MAAM,YAAY;AA/FtC;AAgGY,gBAAI,KAAK,SAAS,gBAAgB,KAAK,MAAM,SAAS,GAAG;AACvD,wBAAU;AAAA,YACZ;AAEA,gBACE,KAAK,SAAS,YACd,4CAA4C;AAAA,cAC1C,iBAAiB,KAAK,aAAa;AAAA,cACnC;AAAA,cACA,qBAAoB,aAAQ,mBAAR,mBAAwB;AAAA,YAC9C,CAAC,GACD;AACA,yBAAW,QAAQ;AAAA,gBACjB,GAAG;AAAA,gBACH,cAAc;AAAA,kBACZ,GAAG,KAAK;AAAA,kBACR,SAAS;AAAA,gBACX;AAAA,cACF,CAAC;AACD;AAAA,YACF;AAEA,kBACE,aAAQ,mBAAR,mBAAwB,UAAS,UACjC,YACC,KAAK,SAAS,sBACb,KAAK,SAAS,sBACd,KAAK,SAAS,oBACd,KAAK,SAAS,cAChB;AACA;AAAA,YACF;AAEA,uBAAW,QAAQ,IAAI;AAAA,UACzB;AAAA,QACF,CAAC;AAAA,MACH;AAAA,IACF;AAAA,EACF;AACF;;;ADzHA,SAAS,SAAS;;;AEZX,IAAM,UACX,OACI,WACA;;;AFaN,IAAM,sBAAsB,EAAE,OAAO;AAAA,EACnC,SAAS,EAAE,OAAO;AAAA,EAClB,MAAM,EAAE,OAAO;AAAA,EACf,OAAO,EAAE,OAAO;AAAA,EAChB,MAAM,EAAE,OAAO;AACjB,CAAC;AAID,IAAM,yBAAoE;AAAA,EACxE,aAAa;AAAA,EACb,gBAAgB,UAAQ,KAAK;AAC/B;AAMA,SAAS,6BACP,MACqB;AACrB,QAAM;AAAA,IACJ,YAAY;AAAA,IACZ;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,GAAG;AAAA,EACL,IAAI;AAEJ,SAAO;AAAA,IACL,GAAG;AAAA,IACH,GAAI,cAAc,UAAa;AAAA,MAC7B,uBAAuB;AAAA,IACzB;AAAA,IACA,GAAI,sBAAsB,UAAa;AAAA,MACrC,qBAAqB;AAAA,IACvB;AAAA,IACA,GAAI,gBAAgB,UAAa,EAAE,cAAc,YAAY;AAAA,IAC7D,GAAI,cAAc,UAAa,EAAE,YAAY,UAAU;AAAA,IACvD,GAAI,gBAAgB,UAAa,EAAE,cAAc,YAAY;AAAA,IAC7D,GAAI,oBAAoB,UAAa;AAAA,MACnC,kBAAkB;AAAA,IACpB;AAAA,IACA,GAAI,mBAAmB,UAAa;AAAA,MAClC,kBAAkB;AAAA,IACpB;AAAA,IACA,UAAU,MAAM,QAAQ,KAAK,QAAQ,IACjC,KAAK,SAAS,IAAI,aAAW;AAC3B,UACE,WAAW,QACX,OAAO,YAAY,YACnB,QAAQ,SAAS,eACjB,EAAE,uBAAuB,UACzB;AACA,eAAO;AAAA,MACT;AAEA,YAAM,EAAE,mBAAmB,GAAG,KAAK,IAAI;AAEvC,aAAO;AAAA,QACL,GAAG;AAAA,QACH,GAAI,EAAE,eAAe,SAAS,sBAAsB,SAChD,EAAE,WAAW,kBAAkB,IAC/B,CAAC;AAAA,MACP;AAAA,IACF,CAAC,IACD,KAAK;AAAA,EACX;AACF;AA4CO,SAAS,eACd,UAAoC,CAAC,GACnB;AAvIpB;AAwIE,QAAM,UAAU;AAAA,KACd,aAAQ,YAAR,YAAmB;AAAA,EACrB;AACA,QAAM,aAAa,MACjB;AAAA,IACE;AAAA,MACE,eAAe,UAAU,WAAW;AAAA,QAClC,QAAQ,QAAQ;AAAA,QAChB,yBAAyB;AAAA,QACzB,aAAa;AAAA,MACf,CAAC,CAAC;AAAA,MACF,GAAG,QAAQ;AAAA,IACb;AAAA,IACA,mBAAmB,OAAO;AAAA,EAC5B;AAEF,QAAM,sBAAsB,CAAC,YAAiC;AAC5D,WAAO,IAAI,0BAA0B,SAAS;AAAA,MAC5C,UAAU;AAAA,MACV,KAAK,CAAC,EAAE,KAAK,MAAM,GAAG,OAAO,GAAG,IAAI;AAAA,MACpC,SAAS;AAAA,MACT,OAAO,QAAQ;AAAA,MACf,gBAAgB;AAAA,MAChB,2BAA2B;AAAA,MAC3B,sBAAsB;AAAA,IACxB,CAAC;AAAA,EACH;AAEA,QAAM,WAAW,CAAC,YAChB,oBAAoB,OAAO;AAE7B,WAAS,uBAAuB;AAChC,WAAS,gBAAgB;AACzB,WAAS,OAAO;AAEhB,WAAS,iBAAiB,CAAC,YAAoB;AAC7C,UAAM,IAAI,iBAAiB,EAAE,SAAS,WAAW,iBAAiB,CAAC;AAAA,EACrE;AACA,WAAS,qBAAqB,SAAS;AACvC,WAAS,aAAa,CAAC,YAAoB;AACzC,UAAM,IAAI,iBAAiB,EAAE,SAAS,WAAW,aAAa,CAAC;AAAA,EACjE;AAEA,SAAO;AACT;AAEO,IAAM,WAAW,eAAe;","names":[]}

package/docs/40-cerebras.mdx

@@ -118,9 +118,41 @@ See [AI SDK UI: Chatbot](/docs/ai-sdk-ui/chatbot#reasoning) for more details on
 
 The following optional provider options are available for Cerebras language models:
 
-- **reasoningEffort** _'low' | 'medium' | 'high'_
+- **parallelToolCalls** _boolean_
 
-  Controls the depth of reasoning for GPT-OSS models. Defaults to `'medium'`.
+  Whether to enable parallel function calling during tool use. Defaults to `true`.
+
+- **logprobs** _boolean_
+
+  Whether to return log probabilities for generated tokens. Defaults to `false`.
+
+- **topLogprobs** _number_
+
+  Number of most likely tokens to return at each token position. Accepts values from `0` through `20`. Requires `logprobs` to be `true`.
+
+- **logitBias** _Record<string, number>_
+
+  Maps token IDs to bias values from `-100` through `100`.
+
+- **serviceTier** _'auto' | 'default' | 'flex' | 'priority'_
+
+  Controls request priority. Availability depends on your account and endpoint.
+
+- **reasoningEffort** _'none' | 'low' | 'medium' | 'high'_
+
+  Controls the amount of reasoning performed by supported models. Supported values and defaults depend on the model.
+
+- **reasoningFormat** _'none' | 'parsed' | 'text_parsed' | 'raw' | 'hidden'_
+
+  Controls how reasoning content appears in the response. Format support depends on the model.
+
+- **prediction** _{ type: 'content'; content: string | Array<{ type: 'text'; text: string }> }_
+
+  Supplies predicted output that can accelerate requests where most of the response is already known.
+
+- **promptCacheKey** _string_
+
+  Routes related requests to the same prompt cache. The value can contain up to 1024 characters and requires account-level enablement.
 
 - **user** _string_
 
@@ -130,6 +162,28 @@ The following optional provider options are available for Cerebras language mode
 
   Whether to use strict JSON schema validation. When `true`, the model uses constrained decoding to guarantee schema compliance. Defaults to `true`.
 
+For example:
+
+```ts
+import {
+  cerebras,
+  type CerebrasLanguageModelChatOptions,
+} from '@ai-sdk/cerebras';
+import { generateText } from 'ai';
+
+const result = await generateText({
+  model: cerebras('gpt-oss-120b'),
+  prompt: 'Explain why the sky is blue.',
+  providerOptions: {
+    cerebras: {
+      reasoningEffort: 'low',
+      reasoningFormat: 'parsed',
+      promptCacheKey: 'conversation-123',
+    } satisfies CerebrasLanguageModelChatOptions,
+  },
+});
+```
+
 ## Model Capabilities
 
 | Model          | Image Input | Object Generation | Tool Usage | Tool Streaming | Reasoning |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-sdk/cerebras",
-  "version": "3.0.32",
+  "version": "3.0.34",
   "type": "module",
   "license": "Apache-2.0",
   "sideEffects": false,
@@ -29,7 +29,7 @@
     }
   },
   "dependencies": {
-    "@ai-sdk/openai-compatible": "3.0.32",
+    "@ai-sdk/openai-compatible": "3.0.34",
     "@ai-sdk/provider": "4.0.7",
     "@ai-sdk/provider-utils": "5.0.28"
   },

package/src/cerebras-chat-language-model-options.ts

@@ -1,3 +1,31 @@
+import { z } from 'zod/v4';
+
 export type { CerebrasChatModelId } from './cerebras-chat-options';
 
-export type CerebrasLanguageModelChatOptions = Record<string, never>;
+export const cerebrasLanguageModelChatOptions = z.object({
+  user: z.string().optional(),
+  strictJsonSchema: z.boolean().optional(),
+  parallelToolCalls: z.boolean().optional(),
+  logprobs: z.boolean().optional(),
+  topLogprobs: z.number().int().min(0).max(20).optional(),
+  logitBias: z.record(z.string(), z.number().min(-100).max(100)).optional(),
+  serviceTier: z.enum(['auto', 'default', 'flex', 'priority']).optional(),
+  reasoningEffort: z.enum(['none', 'low', 'medium', 'high']).optional(),
+  reasoningFormat: z
+    .enum(['none', 'parsed', 'text_parsed', 'raw', 'hidden'])
+    .optional(),
+  prediction: z
+    .object({
+      type: z.literal('content'),
+      content: z.union([
+        z.string(),
+        z.array(z.object({ type: z.literal('text'), text: z.string() })),
+      ]),
+    })
+    .optional(),
+  promptCacheKey: z.string().max(1024).optional(),
+});
+
+export type CerebrasLanguageModelChatOptions = z.infer<
+  typeof cerebrasLanguageModelChatOptions
+>;

package/src/cerebras-provider.ts

@@ -37,8 +37,34 @@ const cerebrasErrorStructure: ProviderErrorStructure<CerebrasErrorData> = {
 function transformCerebrasRequestBody(
   args: Record<string, any>,
 ): Record<string, any> {
+  const {
+    max_tokens: maxTokens,
+    parallelToolCalls,
+    topLogprobs,
+    logitBias,
+    serviceTier,
+    reasoningFormat,
+    promptCacheKey,
+    ...restArgs
+  } = args;
+
   return {
-    ...args,
+    ...restArgs,
+    ...(maxTokens !== undefined && {
+      max_completion_tokens: maxTokens,
+    }),
+    ...(parallelToolCalls !== undefined && {
+      parallel_tool_calls: parallelToolCalls,
+    }),
+    ...(topLogprobs !== undefined && { top_logprobs: topLogprobs }),
+    ...(logitBias !== undefined && { logit_bias: logitBias }),
+    ...(serviceTier !== undefined && { service_tier: serviceTier }),
+    ...(reasoningFormat !== undefined && {
+      reasoning_format: reasoningFormat,
+    }),
+    ...(promptCacheKey !== undefined && {
+      prompt_cache_key: promptCacheKey,
+    }),
     messages: Array.isArray(args.messages)
       ? args.messages.map(message => {
           if (

package/src/index.ts

@@ -4,4 +4,5 @@ export type {
   CerebrasProviderSettings,
 } from './cerebras-provider';
 export type { CerebrasErrorData } from './cerebras-provider';
+export type { CerebrasLanguageModelChatOptions } from './cerebras-chat-language-model-options';
 export { VERSION } from './version';