@ai-sdk/provider 2.0.0-canary.1 → 2.0.0-canary.11

package/dist/index.d.ts

@@ -1,11 +1,65 @@
 import { JSONSchema7 } from 'json-schema';
 export { JSONSchema7, JSONSchema7Definition } from 'json-schema';
 
+type SharedV2Headers = Record<string, string>;
+
+type JSONValue = null | string | number | boolean | JSONObject | JSONArray;
+type JSONObject = {
+    [key: string]: JSONValue;
+};
+type JSONArray = JSONValue[];
+
+/**
+ * Additional provider-specific metadata.
+ * Metadata are additional outputs from the provider.
+ * They are passed through to the provider from the AI SDK
+ * and enable provider-specific functionality
+ * that can be fully encapsulated in the provider.
+ *
+ * This enables us to quickly ship provider-specific functionality
+ * without affecting the core AI SDK.
+ *
+ * The outer record is keyed by the provider name, and the inner
+ * record is keyed by the provider-specific metadata key.
+ *
+ * ```ts
+ * {
+ *   "anthropic": {
+ *     "cacheControl": { "type": "ephemeral" }
+ *   }
+ * }
+ * ```
+ */
+type SharedV2ProviderMetadata = Record<string, Record<string, JSONValue>>;
+
+/**
+ * Additional provider-specific options.
+ * Options are additional input to the provider.
+ * They are passed through to the provider from the AI SDK
+ * and enable provider-specific functionality
+ * that can be fully encapsulated in the provider.
+ *
+ * This enables us to quickly ship provider-specific functionality
+ * without affecting the core AI SDK.
+ *
+ * The outer record is keyed by the provider name, and the inner
+ * record is keyed by the provider-specific metadata key.
+ *
+ * ```ts
+ * {
+ *   "anthropic": {
+ *     "cacheControl": { "type": "ephemeral" }
+ *   }
+ * }
+ * ```
+ */
+type SharedV2ProviderOptions = Record<string, Record<string, JSONValue>>;
+
 /**
 An embedding is a vector, i.e. an array of numbers.
 It is e.g. used to represent a text as a vector of word embeddings.
  */
-type EmbeddingModelV1Embedding = Array<number>;
+type EmbeddingModelV2Embedding = Array<number>;
 
 /**
 Specification for an embedding model that implements the embedding model
@@ -15,7 +69,7 @@ VALUE is the type of the values that the model can embed.
 This will allow us to go beyond text embeddings in the future,
 e.g. to support image embeddings
  */
-type EmbeddingModelV1<VALUE> = {
+type EmbeddingModelV2<VALUE> = {
     /**
   The embedding model must specify which embedding model interface
   version it implements. This will allow us to evolve the embedding
@@ -23,7 +77,7 @@ type EmbeddingModelV1<VALUE> = {
   implementation versions can be handled as a discriminated union
   on our side.
      */
-    readonly specificationVersion: 'v1';
+    readonly specificationVersion: 'v2';
     /**
   Name of the provider for logging purposes.
      */
@@ -34,12 +88,14 @@ type EmbeddingModelV1<VALUE> = {
     readonly modelId: string;
     /**
   Limit of how many embeddings can be generated in a single API call.
+  
+  Use Infinity for models that do not have a limit.
      */
-    readonly maxEmbeddingsPerCall: number | undefined;
+    readonly maxEmbeddingsPerCall: PromiseLike<number | undefined> | number | undefined;
     /**
   True if the model can handle multiple embedding calls in parallel.
      */
-    readonly supportsParallelCalls: boolean;
+    readonly supportsParallelCalls: PromiseLike<boolean> | boolean;
     /**
   Generates a list of embeddings for the given input text.
   
@@ -56,6 +112,12 @@ type EmbeddingModelV1<VALUE> = {
          */
         abortSignal?: AbortSignal;
         /**
+    Additional provider-specific options. They are passed through
+    to the provider from the AI SDK and enable provider-specific
+    functionality that can be fully encapsulated in the provider.
+        */
+        providerOptions?: SharedV2ProviderOptions;
+        /**
       Additional HTTP headers to be sent with the request.
       Only applicable for HTTP-based providers.
          */
@@ -64,7 +126,7 @@ type EmbeddingModelV1<VALUE> = {
         /**
     Generated embeddings. They are in the same order as the input values.
          */
-        embeddings: Array<EmbeddingModelV1Embedding>;
+        embeddings: Array<EmbeddingModelV2Embedding>;
         /**
     Token usage. We only have input tokens for embeddings.
         */
@@ -72,13 +134,17 @@ type EmbeddingModelV1<VALUE> = {
             tokens: number;
         };
         /**
-    Optional raw response information for debugging purposes.
+    Optional response information for debugging purposes.
          */
-        rawResponse?: {
+        response?: {
             /**
       Response headers.
              */
-            headers?: Record<string, string>;
+            headers?: SharedV2Headers;
+            /**
+            The response body.
+            */
+            body?: unknown;
         };
     }>;
 };
@@ -305,13 +371,7 @@ declare class UnsupportedFunctionalityError extends AISDKError {
     static isInstance(error: unknown): error is UnsupportedFunctionalityError;
 }
 
-type JSONValue = null | string | number | boolean | JSONObject | JSONArray;
-type JSONObject = {
-    [key: string]: JSONValue;
-};
-type JSONArray = JSONValue[];
-
-type ImageModelV1CallOptions = {
+type ImageModelV2CallOptions = {
     /**
   Prompt for the image generation.
        */
@@ -367,9 +427,9 @@ type ImageModelV1CallOptions = {
 Warning from the model provider for this call. The call will proceed, but e.g.
 some settings might not be supported, which can lead to suboptimal results.
  */
-type ImageModelV1CallWarning = {
+type ImageModelV2CallWarning = {
     type: 'unsupported-setting';
-    setting: keyof ImageModelV1CallOptions;
+    setting: keyof ImageModelV2CallOptions;
     details?: string;
 } | {
     type: 'other';
@@ -379,7 +439,7 @@ type ImageModelV1CallWarning = {
 /**
 Image generation model specification version 1.
  */
-type ImageModelV1 = {
+type ImageModelV2 = {
     /**
   The image model must specify which image model interface
   version it implements. This will allow us to evolve the image
@@ -404,7 +464,7 @@ type ImageModelV1 = {
     /**
   Generates an array of images.
      */
-    doGenerate(options: ImageModelV1CallOptions): PromiseLike<{
+    doGenerate(options: ImageModelV2CallOptions): PromiseLike<{
         /**
     Generated images as base64 encoded strings or binary data.
     The images should be returned without any unnecessary conversion.
@@ -416,7 +476,7 @@ type ImageModelV1 = {
         /**
     Warnings for the call, e.g. unsupported settings.
          */
-        warnings: Array<ImageModelV1CallWarning>;
+        warnings: Array<ImageModelV2CallWarning>;
         /**
     Response information for telemetry and debugging purposes.
          */
@@ -441,138 +501,13 @@ declare function isJSONValue(value: unknown): value is JSONValue;
 declare function isJSONArray(value: unknown): value is JSONArray;
 declare function isJSONObject(value: unknown): value is JSONObject;
 
-/**
- * Additional provider-specific metadata. They are passed through
- * to the provider from the AI SDK and enable provider-specific
- * functionality that can be fully encapsulated in the provider.
- *
- * This enables us to quickly ship provider-specific functionality
- * without affecting the core AI SDK.
- *
- * The outer record is keyed by the provider name, and the inner
- * record is keyed by the provider-specific metadata key.
- *
- * ```ts
- * {
- *   "anthropic": {
- *     "cacheControl": { "type": "ephemeral" }
- *   }
- * }
- * ```
- */
-type LanguageModelV1ProviderMetadata = Record<string, Record<string, JSONValue>>;
-
-/**
- * A source that has been used as input to generate the response.
- */
-type LanguageModelV1Source = {
-    /**
-     * A URL source. This is return by web search RAG models.
-     */
-    sourceType: 'url';
-    /**
-     * The ID of the source.
-     */
-    id: string;
-    /**
-     * The URL of the source.
-     */
-    url: string;
-    /**
-     * The title of the source.
-     */
-    title?: string;
-    /**
-     * Additional provider metadata for the source.
-     */
-    providerMetadata?: LanguageModelV1ProviderMetadata;
-};
-
-type LanguageModelV1CallSettings = {
-    /**
-  Maximum number of tokens to generate.
-     */
-    maxTokens?: number;
-    /**
-  Temperature setting.
-  
-  It is recommended to set either `temperature` or `topP`, but not both.
-     */
-    temperature?: number;
-    /**
-  Stop sequences.
-  If set, the model will stop generating text when one of the stop sequences is generated.
-  Providers may have limits on the number of stop sequences.
-     */
-    stopSequences?: string[];
-    /**
-  Nucleus sampling.
-  
-  It is recommended to set either `temperature` or `topP`, but not both.
-     */
-    topP?: number;
-    /**
-  Only sample from the top K options for each subsequent token.
-  
-  Used to remove "long tail" low probability responses.
-  Recommended for advanced use cases only. You usually only need to use temperature.
-     */
-    topK?: number;
-    /**
-  Presence penalty setting. It affects the likelihood of the model to
-  repeat information that is already in the prompt.
-     */
-    presencePenalty?: number;
-    /**
-  Frequency penalty setting. It affects the likelihood of the model
-  to repeatedly use the same words or phrases.
-     */
-    frequencyPenalty?: number;
-    /**
-  Response format. The output can either be text or JSON. Default is text.
-  
-  If JSON is selected, a schema can optionally be provided to guide the LLM.
-     */
-    responseFormat?: {
-        type: 'text';
-    } | {
-        type: 'json';
-        /**
-         * JSON schema that the generated output should conform to.
-         */
-        schema?: JSONSchema7;
-        /**
-         * Name of output that should be generated. Used by some providers for additional LLM guidance.
-         */
-        name?: string;
-        /**
-         * Description of the output that should be generated. Used by some providers for additional LLM guidance.
-         */
-        description?: string;
-    };
-    /**
-  The seed (integer) to use for random sampling. If set and supported
-  by the model, calls will generate deterministic results.
-     */
-    seed?: number;
-    /**
-  Abort signal for cancelling the operation.
-     */
-    abortSignal?: AbortSignal;
-    /**
-  Additional HTTP headers to be sent with the request.
-  Only applicable for HTTP-based providers.
-     */
-    headers?: Record<string, string | undefined>;
-};
-
 /**
 A tool has a name, a description, and a set of parameters.
 
 Note: this is **not** the user-facing tool definition. The AI SDK methods will
 map the user-facing tool definitions to this format.
  */
-type LanguageModelV1FunctionTool = {
+type LanguageModelV2FunctionTool = {
     /**
   The type of the tool (always 'function').
      */
@@ -593,6 +528,11 @@ type LanguageModelV1FunctionTool = {
     parameters: JSONSchema7;
 };
 
+/**
+Data content. Can be a Uint8Array, base64 encoded data as a string or a URL.
+*/
+type LanguageModelV2DataContent = Uint8Array | string | URL;
+
 /**
 A prompt is a list of messages.
 
@@ -602,128 +542,92 @@ tool calls. The validation happens at runtime.
 Note: This is not a user-facing prompt. The AI SDK methods will map the
 user-facing prompt types such as chat or instruction prompts to this format.
  */
-type LanguageModelV1Prompt = Array<LanguageModelV1Message>;
-type LanguageModelV1Message = ({
+type LanguageModelV2Prompt = Array<LanguageModelV2Message>;
+type LanguageModelV2Message = ({
     role: 'system';
     content: string;
 } | {
     role: 'user';
-    content: Array<LanguageModelV1TextPart | LanguageModelV1ImagePart | LanguageModelV1FilePart>;
+    content: Array<LanguageModelV2TextPart | LanguageModelV2FilePart>;
 } | {
     role: 'assistant';
-    content: Array<LanguageModelV1TextPart | LanguageModelV1FilePart | LanguageModelV1ReasoningPart | LanguageModelV1RedactedReasoningPart | LanguageModelV1ToolCallPart>;
+    content: Array<LanguageModelV2TextPart | LanguageModelV2FilePart | LanguageModelV2ReasoningPart | LanguageModelV2ToolCallPart>;
 } | {
     role: 'tool';
-    content: Array<LanguageModelV1ToolResultPart>;
+    content: Array<LanguageModelV2ToolResultPart>;
 }) & {
     /**
-     * Additional provider-specific metadata. They are passed through
+     * Additional provider-specific options. They are passed through
      * to the provider from the AI SDK and enable provider-specific
      * functionality that can be fully encapsulated in the provider.
      */
-    providerMetadata?: LanguageModelV1ProviderMetadata;
+    providerOptions?: SharedV2ProviderOptions;
 };
 /**
 Text content part of a prompt. It contains a string of text.
  */
-interface LanguageModelV1TextPart {
+interface LanguageModelV2TextPart {
     type: 'text';
     /**
   The text content.
      */
     text: string;
     /**
-     * Additional provider-specific metadata. They are passed through
+     * Additional provider-specific options. They are passed through
      * to the provider from the AI SDK and enable provider-specific
      * functionality that can be fully encapsulated in the provider.
      */
-    providerMetadata?: LanguageModelV1ProviderMetadata;
+    providerOptions?: SharedV2ProviderOptions;
 }
 /**
 Reasoning content part of a prompt. It contains a string of reasoning text.
  */
-interface LanguageModelV1ReasoningPart {
+interface LanguageModelV2ReasoningPart {
     type: 'reasoning';
     /**
   The reasoning text.
      */
     text: string;
     /**
-  An optional signature for verifying that the reasoning originated from the model.
-     */
-    signature?: string;
-    /**
-  Additional provider-specific metadata. They are passed through
-  to the provider from the AI SDK and enable provider-specific
-  functionality that can be fully encapsulated in the provider.
+     * Additional provider-specific options. They are passed through
+     * to the provider from the AI SDK and enable provider-specific
+     * functionality that can be fully encapsulated in the provider.
      */
-    providerMetadata?: LanguageModelV1ProviderMetadata;
+    providerOptions?: SharedV2ProviderOptions;
 }
 /**
-Redacted reasoning content part of a prompt.
+File content part of a prompt. It contains a file.
  */
-interface LanguageModelV1RedactedReasoningPart {
-    type: 'redacted-reasoning';
+interface LanguageModelV2FilePart {
+    type: 'file';
     /**
-  Redacted reasoning data.
+     * Optional filename of the file.
      */
-    data: string;
+    filename?: string;
     /**
-  Additional provider-specific metadata. They are passed through
-  to the provider from the AI SDK and enable provider-specific
-  functionality that can be fully encapsulated in the provider.
+  File data. Can be a Uint8Array, base64 encoded data as a string or a URL.
+  */
+    data: LanguageModelV2DataContent;
+    /**
+  IANA media type of the file.
+  
+  Can support wildcards, e.g. `image/*` (in which case the provider needs to take appropriate action).
+  
+  @see https://www.iana.org/assignments/media-types/media-types.xhtml
      */
-    providerMetadata?: LanguageModelV1ProviderMetadata;
+    mediaType: string;
+    /**
+     * Additional provider-specific options. They are passed through
+     * to the provider from the AI SDK and enable provider-specific
+     * functionality that can be fully encapsulated in the provider.
+     */
+    providerOptions?: SharedV2ProviderOptions;
 }
 /**
-Image content part of a prompt. It contains an image.
+Tool call content part of a prompt. It contains a tool call (usually generated by the AI model).
  */
-interface LanguageModelV1ImagePart {
-    type: 'image';
-    /**
-  Image data as a Uint8Array (e.g. from a Blob or Buffer) or a URL.
-     */
-    image: Uint8Array | URL;
-    /**
-  Optional mime type of the image.
-     */
-    mimeType?: string;
-    /**
-     * Additional provider-specific metadata. They are passed through
-     * to the provider from the AI SDK and enable provider-specific
-     * functionality that can be fully encapsulated in the provider.
-     */
-    providerMetadata?: LanguageModelV1ProviderMetadata;
-}
-/**
-File content part of a prompt. It contains a file.
- */
-interface LanguageModelV1FilePart {
-    type: 'file';
-    /**
-     * Optional filename of the file.
-     */
-    filename?: string;
-    /**
-  File data as base64 encoded string or as a URL.
-     */
-    data: string | URL;
-    /**
-  Mime type of the file.
-     */
-    mimeType: string;
-    /**
-     * Additional provider-specific metadata. They are passed through
-     * to the provider from the AI SDK and enable provider-specific
-     * functionality that can be fully encapsulated in the provider.
-     */
-    providerMetadata?: LanguageModelV1ProviderMetadata;
-}
-/**
-Tool call content part of a prompt. It contains a tool call (usually generated by the AI model).
- */
-interface LanguageModelV1ToolCallPart {
-    type: 'tool-call';
+interface LanguageModelV2ToolCallPart {
+    type: 'tool-call';
     /**
   ID of the tool call. This ID is used to match the tool call with the tool result.
    */
@@ -737,16 +641,16 @@ interface LanguageModelV1ToolCallPart {
      */
     args: unknown;
     /**
-     * Additional provider-specific metadata. They are passed through
+     * Additional provider-specific options. They are passed through
      * to the provider from the AI SDK and enable provider-specific
      * functionality that can be fully encapsulated in the provider.
      */
-    providerMetadata?: LanguageModelV1ProviderMetadata;
+    providerOptions?: SharedV2ProviderOptions;
 }
 /**
 Tool result content part of a prompt. It contains the result of the tool call with the matching ID.
  */
-interface LanguageModelV1ToolResultPart {
+interface LanguageModelV2ToolResultPart {
     type: 'tool-result';
     /**
   ID of the tool call that this result is associated with.
@@ -781,22 +685,24 @@ base-64 encoded image data
          */
         data: string;
         /**
-Mime type of the image.
+IANA media type of the image.
+
+@see https://www.iana.org/assignments/media-types/media-types.xhtml
          */
-        mimeType?: string;
+        mediaType?: string;
     }>;
     /**
-     * Additional provider-specific metadata. They are passed through
+     * Additional provider-specific options. They are passed through
      * to the provider from the AI SDK and enable provider-specific
      * functionality that can be fully encapsulated in the provider.
      */
-    providerMetadata?: LanguageModelV1ProviderMetadata;
+    providerOptions?: SharedV2ProviderOptions;
 }
 
 /**
 The configuration of a tool that is defined by the provider.
  */
-type LanguageModelV1ProviderDefinedTool = {
+type LanguageModelV2ProviderDefinedTool = {
     /**
   The type of the tool (always 'provider-defined').
      */
@@ -815,7 +721,7 @@ type LanguageModelV1ProviderDefinedTool = {
     args: Record<string, unknown>;
 };
 
-type LanguageModelV1ToolChoice = {
+type LanguageModelV2ToolChoice = {
     type: 'auto';
 } | {
     type: 'none';
@@ -826,38 +732,64 @@ type LanguageModelV1ToolChoice = {
     toolName: string;
 };
 
-type LanguageModelV1CallOptions = LanguageModelV1CallSettings & {
+type LanguageModelV2CallOptions = {
     /**
-  Whether the user provided the input as messages or as
-  a prompt. This can help guide non-chat models in the
-  expansion, bc different expansions can be needed for
-  chat/non-chat use cases.
+  A language mode prompt is a standardized prompt type.
+  
+  Note: This is **not** the user-facing prompt. The AI SDK methods will map the
+  user-facing prompt types such as chat or instruction prompts to this format.
+  That approach allows us to evolve the user  facing prompts without breaking
+  the language model interface.
      */
-    inputFormat: 'messages' | 'prompt';
+    prompt: LanguageModelV2Prompt;
     /**
-  The mode affects the behavior of the language model. It is required to
-  support provider-independent streaming and generation of structured objects.
-  The model can take this information and e.g. configure json mode, the correct
-  low level grammar, etc. It can also be used to optimize the efficiency of the
-  streaming, e.g. tool-delta stream parts are only needed in the
-  object-tool mode.
-  
-  @deprecated mode will be removed in v2.
-  All necessary settings will be directly supported through the call settings,
-  in particular responseFormat, toolChoice, and tools.
+  Maximum number of tokens to generate.
      */
-    mode: {
-        type: 'regular';
-        /**
-The tools that are available for the model.
-         */
-        tools?: Array<LanguageModelV1FunctionTool | LanguageModelV1ProviderDefinedTool>;
-        /**
-Specifies how the tool should be selected. Defaults to 'auto'.
-         */
-        toolChoice?: LanguageModelV1ToolChoice;
+    maxOutputTokens?: number;
+    /**
+  Temperature setting.
+  
+  It is recommended to set either `temperature` or `topP`, but not both.
+   */
+    temperature?: number;
+    /**
+  Stop sequences.
+  If set, the model will stop generating text when one of the stop sequences is generated.
+  Providers may have limits on the number of stop sequences.
+   */
+    stopSequences?: string[];
+    /**
+  Nucleus sampling.
+  
+  It is recommended to set either `temperature` or `topP`, but not both.
+   */
+    topP?: number;
+    /**
+  Only sample from the top K options for each subsequent token.
+  
+  Used to remove "long tail" low probability responses.
+  Recommended for advanced use cases only. You usually only need to use temperature.
+   */
+    topK?: number;
+    /**
+  Presence penalty setting. It affects the likelihood of the model to
+  repeat information that is already in the prompt.
+   */
+    presencePenalty?: number;
+    /**
+  Frequency penalty setting. It affects the likelihood of the model
+  to repeatedly use the same words or phrases.
+   */
+    frequencyPenalty?: number;
+    /**
+  Response format. The output can either be text or JSON. Default is text.
+  
+  If JSON is selected, a schema can optionally be provided to guide the LLM.
+   */
+    responseFormat?: {
+        type: 'text';
     } | {
-        type: 'object-json';
+        type: 'json';
         /**
          * JSON schema that the generated output should conform to.
          */
@@ -870,38 +802,48 @@ Specifies how the tool should be selected. Defaults to 'auto'.
          * Description of the output that should be generated. Used by some providers for additional LLM guidance.
          */
         description?: string;
-    } | {
-        type: 'object-tool';
-        tool: LanguageModelV1FunctionTool;
     };
     /**
-  A language mode prompt is a standardized prompt type.
-  
-  Note: This is **not** the user-facing prompt. The AI SDK methods will map the
-  user-facing prompt types such as chat or instruction prompts to this format.
-  That approach allows us to evolve the user  facing prompts without breaking
-  the language model interface.
-     */
-    prompt: LanguageModelV1Prompt;
+  The seed (integer) to use for random sampling. If set and supported
+  by the model, calls will generate deterministic results.
+   */
+    seed?: number;
     /**
-  Additional provider-specific metadata.
-  The metadata is passed through to the provider from the AI SDK and enables
-  provider-specific functionality that can be fully encapsulated in the provider.
+  The tools that are available for the model.
+    */
+    tools?: Array<LanguageModelV2FunctionTool | LanguageModelV2ProviderDefinedTool>;
+    /**
+  Specifies how the tool should be selected. Defaults to 'auto'.
+  */
+    toolChoice?: LanguageModelV2ToolChoice;
+    /**
+  Abort signal for cancelling the operation.
+   */
+    abortSignal?: AbortSignal;
+    /**
+  Additional HTTP headers to be sent with the request.
+  Only applicable for HTTP-based providers.
+   */
+    headers?: Record<string, string | undefined>;
+    /**
+     * Additional provider-specific options. They are passed through
+     * to the provider from the AI SDK and enable provider-specific
+     * functionality that can be fully encapsulated in the provider.
      */
-    providerMetadata?: LanguageModelV1ProviderMetadata;
+    providerOptions?: SharedV2ProviderOptions;
 };
 
 /**
 Warning from the model provider for this call. The call will proceed, but e.g.
 some settings might not be supported, which can lead to suboptimal results.
  */
-type LanguageModelV1CallWarning = {
+type LanguageModelV2CallWarning = {
     type: 'unsupported-setting';
-    setting: keyof LanguageModelV1CallSettings;
+    setting: Omit<keyof LanguageModelV2CallOptions, 'prompt'>;
     details?: string;
 } | {
     type: 'unsupported-tool';
-    tool: LanguageModelV1FunctionTool | LanguageModelV1ProviderDefinedTool;
+    tool: LanguageModelV2FunctionTool | LanguageModelV2ProviderDefinedTool;
     details?: string;
 } | {
     type: 'other';
@@ -909,230 +851,247 @@ type LanguageModelV1CallWarning = {
 };
 
 /**
-Reason why a language model finished generating a response.
-
-Can be one of the following:
-- `stop`: model generated stop sequence
-- `length`: model generated maximum number of tokens
-- `content-filter`: content filter violation stopped the model
-- `tool-calls`: model triggered tool calls
-- `error`: model stopped because of an error
-- `other`: model stopped for other reasons
-- `unknown`: the model has not transmitted a finish reason
+A file that has been generated by the model.
+Generated files as base64 encoded strings or binary data.
+The files should be returned without any unnecessary conversion.
  */
-type LanguageModelV1FinishReason = 'stop' | 'length' | 'content-filter' | 'tool-calls' | 'error' | 'other' | 'unknown';
-
-type LanguageModelV1FunctionToolCall = {
-    toolCallType: 'function';
-    toolCallId: string;
-    toolName: string;
+type LanguageModelV2File = {
+    type: 'file';
     /**
-  Stringified JSON object with the tool call arguments. Must match the
-  parameters schema of the tool.
-     */
-    args: string;
+  The IANA media type of the file, e.g. `image/png` or `audio/mp3`.
+  
+  @see https://www.iana.org/assignments/media-types/media-types.xhtml
+         */
+    mediaType: string;
+    /**
+  Generated file data as base64 encoded strings or binary data.
+  
+  The file data should be returned without any unnecessary conversion.
+  If the API returns base64 encoded strings, the file data should be returned
+  as base64 encoded strings. If the API returns binary data, the file data should
+  be returned as binary data.
+   */
+    data: string | Uint8Array;
 };
 
 /**
-Log probabilities for each token and its top log probabilities.
+Reasoning that the model has generated.
  */
-type LanguageModelV1LogProbs = Array<{
-    token: string;
-    logprob: number;
-    topLogprobs: Array<{
-        token: string;
-        logprob: number;
-    }>;
-}>;
+type LanguageModelV2Reasoning = {
+    type: 'reasoning';
+    text: string;
+    /**
+     * Optional provider-specific metadata for the reasoning part.
+     */
+    providerMetadata?: SharedV2ProviderMetadata;
+};
 
 /**
-Specification for a language model that implements the language model interface version 1.
+A source that has been used as input to generate the response.
  */
-type LanguageModelV1 = {
+type LanguageModelV2Source = {
+    type: 'source';
     /**
-  The language model must specify which language model interface
-  version it implements. This will allow us to evolve the language
-  model interface and retain backwards compatibility. The different
-  implementation versions can be handled as a discriminated union
-  on our side.
+     * A URL source. This is return by web search RAG models.
      */
-    readonly specificationVersion: 'v1';
+    sourceType: 'url';
     /**
-  Name of the provider for logging purposes.
+     * The ID of the source.
      */
-    readonly provider: string;
+    id: string;
     /**
-  Provider-specific model ID for logging purposes.
+     * The URL of the source.
      */
-    readonly modelId: string;
+    url: string;
     /**
-  Default object generation mode that should be used with this model when
-  no mode is specified. Should be the mode with the best results for this
-  model. `undefined` can be returned if object generation is not supported.
-  
-  This is needed to generate the best objects possible w/o requiring the
-  user to explicitly specify the object generation mode.
+     * The title of the source.
      */
-    readonly defaultObjectGenerationMode: LanguageModelV1ObjectGenerationMode;
+    title?: string;
     /**
-  Flag whether this model supports image URLs. Default is `true`.
-  
-  When the flag is set to `false`, the AI SDK will download the image and
-  pass the image data to the model.
+     * Additional provider metadata for the source.
      */
-    readonly supportsImageUrls?: boolean;
+    providerMetadata?: SharedV2ProviderMetadata;
+};
+
+/**
+Text that the model has generated.
+ */
+type LanguageModelV2Text = {
+    type: 'text';
     /**
-  Flag whether this model supports grammar-guided generation,
-  i.e. follows JSON schemas for object generation
-  when the response format is set to 'json' or
-  when the `object-json` mode is used.
-  
-  This means that the model guarantees that the generated JSON
-  will be a valid JSON object AND that the object will match the
-  JSON schema.
-  
-  Please note that `generateObject` and `streamObject` will work
-  regardless of this flag, but might send different prompts and
-  use further optimizations if this flag is set to `true`.
-  
-  Defaults to `false`.
+  The text content.
+     */
+    text: string;
+};
+
+/**
+Tool calls that the model has generated.
+     */
+type LanguageModelV2ToolCall = {
+    type: 'tool-call';
+    toolCallType: 'function';
+    toolCallId: string;
+    toolName: string;
+    /**
+  Stringified JSON object with the tool call arguments. Must match the
+  parameters schema of the tool.
+     */
+    args: string;
+};
+
+type LanguageModelV2Content = LanguageModelV2Text | LanguageModelV2Reasoning | LanguageModelV2File | LanguageModelV2Source | LanguageModelV2ToolCall;
+
+/**
+Reason why a language model finished generating a response.
+
+Can be one of the following:
+- `stop`: model generated stop sequence
+- `length`: model generated maximum number of tokens
+- `content-filter`: content filter violation stopped the model
+- `tool-calls`: model triggered tool calls
+- `error`: model stopped because of an error
+- `other`: model stopped for other reasons
+- `unknown`: the model has not transmitted a finish reason
+ */
+type LanguageModelV2FinishReason = 'stop' | 'length' | 'content-filter' | 'tool-calls' | 'error' | 'other' | 'unknown';
+
+interface LanguageModelV2ResponseMetadata {
+    /**
+  ID for the generated response, if the provider sends one.
+       */
+    id?: string;
+    /**
+  Timestamp for the start of the generated response, if the provider sends one.
   */
-    readonly supportsStructuredOutputs?: boolean;
+    timestamp?: Date;
     /**
-  Checks if the model supports the given URL for file parts natively.
-  If the model does not support the URL,
-  the AI SDK will download the file and pass the file data to the model.
-  
-  When undefined, the AI SDK will download the file.
+  The ID of the response model that was used to generate the response, if the provider sends one.
+  */
+    modelId?: string;
+}
+
+type LanguageModelV2ToolCallDelta = {
+    type: 'tool-call-delta';
+    toolCallType: 'function';
+    toolCallId: string;
+    toolName: string;
+    argsTextDelta: string;
+};
+
+/**
+ * Usage information for a language model call.
+ */
+type LanguageModelV2Usage = {
+    /**
+     * The number of input (prompt) tokens used.
      */
-    supportsUrl?(url: URL): boolean;
+    inputTokens: number | undefined;
+    /**
+     * The number of output (completion) tokens used.
+     */
+    outputTokens: number | undefined;
+};
+
+type LanguageModelV2StreamPart = LanguageModelV2Content | {
+    type: 'reasoning-part-finish';
+} | LanguageModelV2ToolCallDelta | {
+    type: 'stream-start';
+    warnings: Array<LanguageModelV2CallWarning>;
+} | ({
+    type: 'response-metadata';
+} & LanguageModelV2ResponseMetadata) | {
+    type: 'finish';
+    usage: LanguageModelV2Usage;
+    finishReason: LanguageModelV2FinishReason;
+    providerMetadata?: SharedV2ProviderMetadata;
+} | {
+    type: 'error';
+    error: unknown;
+};
+
+/**
+Specification for a language model that implements the language model interface version 2.
+ */
+type LanguageModelV2 = {
+    /**
+  The language model must specify which language model interface
+  version it implements. This will allow us to evolve the language
+  model interface and retain backwards compatibility. The different
+  implementation versions can be handled as a discriminated union
+  on our side.
+     */
+    readonly specificationVersion: 'v2';
+    /**
+  Name of the provider for logging purposes.
+     */
+    readonly provider: string;
+    /**
+  Provider-specific model ID for logging purposes.
+     */
+    readonly modelId: string;
+    /**
+     * Returns a map of supported URL patterns for the model.
+     * The keys are media type patterns or full media types (e.g. `*\/*` for everything, `audio/*`, `video/*`, or `application/pdf`).
+     * and the values are arrays of regular expressions that match the URL paths.
+     *
+     * The matching should be against lower-case URLs.
+     *
+     * Matched URLs are supported natively by the model and are not downloaded.
+     *
+     * @returns A promise resolving to a map of supported URL patterns.
+     */
+    getSupportedUrls(): PromiseLike<Record<string, RegExp[]>>;
     /**
   Generates a language model output (non-streaming).
   
   Naming: "do" prefix to prevent accidental direct usage of the method
   by the user.
      */
-    doGenerate(options: LanguageModelV1CallOptions): PromiseLike<{
-        /**
-    Text that the model has generated.
-    Can be undefined if the model did not generate any text.
-         */
-        text?: string;
-        /**
-    Reasoning that the model has generated.
-    Can be undefined if the model does not support reasoning.
-         */
-        reasoning?: string | Array<{
-            type: 'text';
-            text: string;
-            /**
-An optional signature for verifying that the reasoning originated from the model.
- */
-            signature?: string;
-        } | {
-            type: 'redacted';
-            data: string;
-        }>;
+    doGenerate(options: LanguageModelV2CallOptions): PromiseLike<{
         /**
-    Generated files as base64 encoded strings or binary data.
-    The files should be returned without any unnecessary conversion.
-    If the API returns base64 encoded strings, the files should be returned
-    as base64 encoded strings. If the API returns binary data, the files should
-    be returned as binary data.
+    Ordered content that the model has generated.
          */
-        files?: Array<{
-            data: string | Uint8Array;
-            mimeType: string;
-        }>;
-        /**
-    Tool calls that the model has generated.
-    Can be undefined if the model did not generate any tool calls.
-         */
-        toolCalls?: Array<LanguageModelV1FunctionToolCall>;
+        content: Array<LanguageModelV2Content>;
         /**
     Finish reason.
          */
-        finishReason: LanguageModelV1FinishReason;
+        finishReason: LanguageModelV2FinishReason;
         /**
       Usage information.
          */
-        usage: {
-            promptTokens: number;
-            completionTokens: number;
-        };
-        /**
-    Raw prompt and setting information for observability provider integration.
-         */
-        rawCall: {
-            /**
-      Raw prompt after expansion and conversion to the format that the
-      provider uses to send the information to their API.
-             */
-            rawPrompt: unknown;
-            /**
-      Raw settings that are used for the API call. Includes provider-specific
-      settings.
-             */
-            rawSettings: Record<string, unknown>;
-        };
+        usage: LanguageModelV2Usage;
         /**
-    Optional response information for telemetry and debugging purposes.
+    Additional provider-specific metadata. They are passed through
+    from the provider to the AI SDK and enable provider-specific
+    results that can be fully encapsulated in the provider.
          */
-        rawResponse?: {
-            /**
-      Response headers.
-            */
-            headers?: Record<string, string>;
-            /**
-      Response body.
-            */
-            body?: unknown;
-        };
+        providerMetadata?: SharedV2ProviderMetadata;
         /**
     Optional request information for telemetry and debugging purposes.
          */
         request?: {
             /**
-      Raw request HTTP body that was sent to the provider API as a string (JSON should be stringified).
-      Non-HTTP(s) providers should not set this.
+      Request HTTP body that was sent to the provider API.
              */
-            body?: string;
+            body?: unknown;
         };
         /**
     Optional response information for telemetry and debugging purposes.
          */
-        response?: {
-            /**
-      ID for the generated response, if the provider sends one.
-           */
-            id?: string;
+        response?: LanguageModelV2ResponseMetadata & {
             /**
-      Timestamp for the start of the generated response, if the provider sends one.
-           */
-            timestamp?: Date;
+      Response headers.
+            */
+            headers?: SharedV2Headers;
             /**
-      The ID of the response model that was used to generate the response, if the provider sends one.
-           */
-            modelId?: string;
+      Response HTTP body.
+      */
+            body?: unknown;
         };
-        warnings?: LanguageModelV1CallWarning[];
-        /**
-    Additional provider-specific metadata. They are passed through
-    from the provider to the AI SDK and enable provider-specific
-    results that can be fully encapsulated in the provider.
-         */
-        providerMetadata?: LanguageModelV1ProviderMetadata;
         /**
-    Sources that have been used as input to generate the response.
-         */
-        sources?: LanguageModelV1Source[];
-        /**
-    Logprobs for the completion.
-    `undefined` if the mode does not support logprobs or if was not enabled
-    
-    @deprecated will be changed into a provider-specific extension in v2
+    Warnings for the call, e.g. unsupported settings.
          */
-        logprobs?: LanguageModelV1LogProbs;
+        warnings: Array<LanguageModelV2CallWarning>;
     }>;
     /**
   Generates a language model output (streaming).
@@ -1142,105 +1101,209 @@ An optional signature for verifying that the reasoning originated from the model
      *
   @return A stream of higher-level language model output parts.
      */
-    doStream(options: LanguageModelV1CallOptions): PromiseLike<{
-        stream: ReadableStream<LanguageModelV1StreamPart>;
+    doStream(options: LanguageModelV2CallOptions): PromiseLike<{
+        stream: ReadableStream<LanguageModelV2StreamPart>;
         /**
-    Raw prompt and setting information for observability provider integration.
+    Optional request information for telemetry and debugging purposes.
          */
-        rawCall: {
-            /**
-      Raw prompt after expansion and conversion to the format that the
-      provider uses to send the information to their API.
-             */
-            rawPrompt: unknown;
+        request?: {
             /**
-      Raw settings that are used for the API call. Includes provider-specific
-      settings.
-             */
-            rawSettings: Record<string, unknown>;
+      Request HTTP body that was sent to the provider API.
+         */
+            body?: unknown;
         };
         /**
-    Optional raw response data.
+    Optional response data.
          */
-        rawResponse?: {
+        response?: {
             /**
       Response headers.
              */
-            headers?: Record<string, string>;
+            headers?: SharedV2Headers;
         };
+    }>;
+};
+
+/**
+ * Experimental middleware for LanguageModelV2.
+ * This type defines the structure for middleware that can be used to modify
+ * the behavior of LanguageModelV2 operations.
+ */
+type LanguageModelV2Middleware = {
+    /**
+     * Middleware specification version. Use `v2` for the current version.
+     */
+    middlewareVersion?: 'v2' | undefined;
+    /**
+     * Transforms the parameters before they are passed to the language model.
+     * @param options - Object containing the type of operation and the parameters.
+     * @param options.type - The type of operation ('generate' or 'stream').
+     * @param options.params - The original parameters for the language model call.
+     * @returns A promise that resolves to the transformed parameters.
+     */
+    transformParams?: (options: {
+        type: 'generate' | 'stream';
+        params: LanguageModelV2CallOptions;
+    }) => PromiseLike<LanguageModelV2CallOptions>;
+    /**
+     * Wraps the generate operation of the language model.
+     * @param options - Object containing the generate function, parameters, and model.
+     * @param options.doGenerate - The original generate function.
+     * @param options.doStream - The original stream function.
+     * @param options.params - The parameters for the generate call. If the
+     * `transformParams` middleware is used, this will be the transformed parameters.
+     * @param options.model - The language model instance.
+     * @returns A promise that resolves to the result of the generate operation.
+     */
+    wrapGenerate?: (options: {
+        doGenerate: () => ReturnType<LanguageModelV2['doGenerate']>;
+        doStream: () => ReturnType<LanguageModelV2['doStream']>;
+        params: LanguageModelV2CallOptions;
+        model: LanguageModelV2;
+    }) => Promise<Awaited<ReturnType<LanguageModelV2['doGenerate']>>>;
+    /**
+     * Wraps the stream operation of the language model.
+     *
+     * @param options - Object containing the stream function, parameters, and model.
+     * @param options.doGenerate - The original generate function.
+     * @param options.doStream - The original stream function.
+     * @param options.params - The parameters for the stream call. If the
+     * `transformParams` middleware is used, this will be the transformed parameters.
+     * @param options.model - The language model instance.
+     * @returns A promise that resolves to the result of the stream operation.
+     */
+    wrapStream?: (options: {
+        doGenerate: () => ReturnType<LanguageModelV2['doGenerate']>;
+        doStream: () => ReturnType<LanguageModelV2['doStream']>;
+        params: LanguageModelV2CallOptions;
+        model: LanguageModelV2;
+    }) => PromiseLike<Awaited<ReturnType<LanguageModelV2['doStream']>>>;
+};
+
+/**
+ * Additional provider-specific metadata. They are passed through
+ * to the provider from the AI SDK and enable provider-specific
+ * functionality that can be fully encapsulated in the provider.
+ *
+ * This enables us to quickly ship provider-specific functionality
+ * without affecting the core AI SDK.
+ *
+ * The outer record is keyed by the provider name, and the inner
+ * record is keyed by the provider-specific metadata key.
+ *
+ * ```ts
+ * {
+ *   "anthropic": {
+ *     "cacheControl": { "type": "ephemeral" }
+ *   }
+ * }
+ * ```
+ */
+type LanguageModelV1ProviderMetadata = Record<string, Record<string, JSONValue>>;
+
+/**
+ * A source that has been used as input to generate the response.
+ */
+type LanguageModelV1Source = {
+    /**
+     * A URL source. This is return by web search RAG models.
+     */
+    sourceType: 'url';
+    /**
+     * The ID of the source.
+     */
+    id: string;
+    /**
+     * The URL of the source.
+     */
+    url: string;
+    /**
+     * The title of the source.
+     */
+    title?: string;
+    /**
+     * Additional provider metadata for the source.
+     */
+    providerMetadata?: LanguageModelV1ProviderMetadata;
+};
+
+type LanguageModelV1CallSettings = {
+    /**
+  Maximum number of tokens to generate.
+     */
+    maxTokens?: number;
+    /**
+  Temperature setting.
+  
+  It is recommended to set either `temperature` or `topP`, but not both.
+     */
+    temperature?: number;
+    /**
+  Stop sequences.
+  If set, the model will stop generating text when one of the stop sequences is generated.
+  Providers may have limits on the number of stop sequences.
+     */
+    stopSequences?: string[];
+    /**
+  Nucleus sampling.
+  
+  It is recommended to set either `temperature` or `topP`, but not both.
+     */
+    topP?: number;
+    /**
+  Only sample from the top K options for each subsequent token.
+  
+  Used to remove "long tail" low probability responses.
+  Recommended for advanced use cases only. You usually only need to use temperature.
+     */
+    topK?: number;
+    /**
+  Presence penalty setting. It affects the likelihood of the model to
+  repeat information that is already in the prompt.
+     */
+    presencePenalty?: number;
+    /**
+  Frequency penalty setting. It affects the likelihood of the model
+  to repeatedly use the same words or phrases.
+     */
+    frequencyPenalty?: number;
+    /**
+  Response format. The output can either be text or JSON. Default is text.
+  
+  If JSON is selected, a schema can optionally be provided to guide the LLM.
+     */
+    responseFormat?: {
+        type: 'text';
+    } | {
+        type: 'json';
         /**
-    Optional request information for telemetry and debugging purposes.
+         * JSON schema that the generated output should conform to.
          */
-        request?: {
-            /**
-      Raw request HTTP body that was sent to the provider API as a string (JSON should be stringified).
-      Non-HTTP(s) providers should not set this.
+        schema?: JSONSchema7;
+        /**
+         * Name of output that should be generated. Used by some providers for additional LLM guidance.
          */
-            body?: string;
-        };
+        name?: string;
         /**
-    Warnings for the call, e.g. unsupported settings.
+         * Description of the output that should be generated. Used by some providers for additional LLM guidance.
          */
-        warnings?: Array<LanguageModelV1CallWarning>;
-    }>;
-};
-type LanguageModelV1StreamPart = {
-    type: 'text-delta';
-    textDelta: string;
-} | {
-    type: 'reasoning';
-    textDelta: string;
-} | {
-    type: 'reasoning-signature';
-    signature: string;
-} | {
-    type: 'redacted-reasoning';
-    data: string;
-} | {
-    type: 'source';
-    source: LanguageModelV1Source;
-} | {
-    type: 'file';
-    mimeType: string;
+        description?: string;
+    };
     /**
-Generated file data as base64 encoded strings or binary data.
-The file data should be returned without any unnecessary conversion.
-If the API returns base64 encoded strings, the file data should be returned
-as base64 encoded strings. If the API returns binary data, the file data should
-be returned as binary data.
+  The seed (integer) to use for random sampling. If set and supported
+  by the model, calls will generate deterministic results.
      */
-    data: string | Uint8Array;
-} | ({
-    type: 'tool-call';
-} & LanguageModelV1FunctionToolCall) | {
-    type: 'tool-call-delta';
-    toolCallType: 'function';
-    toolCallId: string;
-    toolName: string;
-    argsTextDelta: string;
-} | {
-    type: 'response-metadata';
-    id?: string;
-    timestamp?: Date;
-    modelId?: string;
-} | {
-    type: 'finish';
-    finishReason: LanguageModelV1FinishReason;
-    providerMetadata?: LanguageModelV1ProviderMetadata;
-    usage: {
-        promptTokens: number;
-        completionTokens: number;
-    };
-    logprobs?: LanguageModelV1LogProbs;
-} | {
-    type: 'error';
-    error: unknown;
-};
-/**
-The object generation modes available for use with a model. `undefined`
-represents no support for object generation.
-   */
-type LanguageModelV1ObjectGenerationMode = 'json' | 'tool' | undefined;
+    seed?: number;
+    /**
+  Abort signal for cancelling the operation.
+     */
+    abortSignal?: AbortSignal;
+    /**
+  Additional HTTP headers to be sent with the request.
+  Only applicable for HTTP-based providers.
+     */
+    headers?: Record<string, string | undefined>;
+};
 
 /**
 A tool has a name, a description, and a set of parameters.
@@ -1248,7 +1311,7 @@ A tool has a name, a description, and a set of parameters.
 Note: this is **not** the user-facing tool definition. The AI SDK methods will
 map the user-facing tool definitions to this format.
  */
-type LanguageModelV2FunctionTool = {
+type LanguageModelV1FunctionTool = {
     /**
   The type of the tool (always 'function').
      */
@@ -1269,29 +1332,6 @@ type LanguageModelV2FunctionTool = {
     parameters: JSONSchema7;
 };
 
-/**
- * Additional provider-specific options.
- * Options are additional input to the provider.
- * They are passed through to the provider from the AI SDK
- * and enable provider-specific functionality
- * that can be fully encapsulated in the provider.
- *
- * This enables us to quickly ship provider-specific functionality
- * without affecting the core AI SDK.
- *
- * The outer record is keyed by the provider name, and the inner
- * record is keyed by the provider-specific metadata key.
- *
- * ```ts
- * {
- *   "anthropic": {
- *     "cacheControl": { "type": "ephemeral" }
- *   }
- * }
- * ```
- */
-type LanguageModelV2ProviderOptions = Record<string, Record<string, JSONValue>>;
-
 /**
 A prompt is a list of messages.
 
@@ -1301,47 +1341,47 @@ tool calls. The validation happens at runtime.
 Note: This is not a user-facing prompt. The AI SDK methods will map the
 user-facing prompt types such as chat or instruction prompts to this format.
  */
-type LanguageModelV2Prompt = Array<LanguageModelV2Message>;
-type LanguageModelV2Message = ({
+type LanguageModelV1Prompt = Array<LanguageModelV1Message>;
+type LanguageModelV1Message = ({
     role: 'system';
     content: string;
 } | {
     role: 'user';
-    content: Array<LanguageModelV2TextPart | LanguageModelV2ImagePart | LanguageModelV2FilePart>;
+    content: Array<LanguageModelV1TextPart | LanguageModelV1ImagePart | LanguageModelV1FilePart>;
 } | {
     role: 'assistant';
-    content: Array<LanguageModelV2TextPart | LanguageModelV2FilePart | LanguageModelV2ReasoningPart | LanguageModelV2RedactedReasoningPart | LanguageModelV2ToolCallPart>;
+    content: Array<LanguageModelV1TextPart | LanguageModelV1FilePart | LanguageModelV1ReasoningPart | LanguageModelV1RedactedReasoningPart | LanguageModelV1ToolCallPart>;
 } | {
     role: 'tool';
-    content: Array<LanguageModelV2ToolResultPart>;
+    content: Array<LanguageModelV1ToolResultPart>;
 }) & {
     /**
-     * Additional provider-specific options. They are passed through
+     * Additional provider-specific metadata. They are passed through
      * to the provider from the AI SDK and enable provider-specific
      * functionality that can be fully encapsulated in the provider.
      */
-    providerOptions?: LanguageModelV2ProviderOptions;
+    providerMetadata?: LanguageModelV1ProviderMetadata;
 };
 /**
 Text content part of a prompt. It contains a string of text.
  */
-interface LanguageModelV2TextPart {
+interface LanguageModelV1TextPart {
     type: 'text';
     /**
   The text content.
      */
     text: string;
     /**
-     * Additional provider-specific options. They are passed through
+     * Additional provider-specific metadata. They are passed through
      * to the provider from the AI SDK and enable provider-specific
      * functionality that can be fully encapsulated in the provider.
      */
-    providerOptions?: LanguageModelV2ProviderOptions;
+    providerMetadata?: LanguageModelV1ProviderMetadata;
 }
 /**
 Reasoning content part of a prompt. It contains a string of reasoning text.
  */
-interface LanguageModelV2ReasoningPart {
+interface LanguageModelV1ReasoningPart {
     type: 'reasoning';
     /**
   The reasoning text.
@@ -1352,32 +1392,32 @@ interface LanguageModelV2ReasoningPart {
      */
     signature?: string;
     /**
-     * Additional provider-specific options. They are passed through
-     * to the provider from the AI SDK and enable provider-specific
-     * functionality that can be fully encapsulated in the provider.
+  Additional provider-specific metadata. They are passed through
+  to the provider from the AI SDK and enable provider-specific
+  functionality that can be fully encapsulated in the provider.
      */
-    providerOptions?: LanguageModelV2ProviderOptions;
+    providerMetadata?: LanguageModelV1ProviderMetadata;
 }
 /**
 Redacted reasoning content part of a prompt.
  */
-interface LanguageModelV2RedactedReasoningPart {
+interface LanguageModelV1RedactedReasoningPart {
     type: 'redacted-reasoning';
     /**
   Redacted reasoning data.
      */
     data: string;
     /**
-     * Additional provider-specific options. They are passed through
-     * to the provider from the AI SDK and enable provider-specific
-     * functionality that can be fully encapsulated in the provider.
+  Additional provider-specific metadata. They are passed through
+  to the provider from the AI SDK and enable provider-specific
+  functionality that can be fully encapsulated in the provider.
      */
-    providerOptions?: LanguageModelV2ProviderOptions;
+    providerMetadata?: LanguageModelV1ProviderMetadata;
 }
 /**
 Image content part of a prompt. It contains an image.
  */
-interface LanguageModelV2ImagePart {
+interface LanguageModelV1ImagePart {
     type: 'image';
     /**
   Image data as a Uint8Array (e.g. from a Blob or Buffer) or a URL.
@@ -1388,16 +1428,16 @@ interface LanguageModelV2ImagePart {
      */
     mimeType?: string;
     /**
-     * Additional provider-specific options. They are passed through
+     * Additional provider-specific metadata. They are passed through
      * to the provider from the AI SDK and enable provider-specific
      * functionality that can be fully encapsulated in the provider.
      */
-    providerOptions?: LanguageModelV2ProviderOptions;
+    providerMetadata?: LanguageModelV1ProviderMetadata;
 }
 /**
 File content part of a prompt. It contains a file.
  */
-interface LanguageModelV2FilePart {
+interface LanguageModelV1FilePart {
     type: 'file';
     /**
      * Optional filename of the file.
@@ -1412,16 +1452,16 @@ interface LanguageModelV2FilePart {
      */
     mimeType: string;
     /**
-     * Additional provider-specific options. They are passed through
+     * Additional provider-specific metadata. They are passed through
      * to the provider from the AI SDK and enable provider-specific
      * functionality that can be fully encapsulated in the provider.
      */
-    providerOptions?: LanguageModelV2ProviderOptions;
+    providerMetadata?: LanguageModelV1ProviderMetadata;
 }
 /**
 Tool call content part of a prompt. It contains a tool call (usually generated by the AI model).
  */
-interface LanguageModelV2ToolCallPart {
+interface LanguageModelV1ToolCallPart {
     type: 'tool-call';
     /**
   ID of the tool call. This ID is used to match the tool call with the tool result.
@@ -1436,16 +1476,16 @@ interface LanguageModelV2ToolCallPart {
      */
     args: unknown;
     /**
-     * Additional provider-specific options. They are passed through
+     * Additional provider-specific metadata. They are passed through
      * to the provider from the AI SDK and enable provider-specific
      * functionality that can be fully encapsulated in the provider.
      */
-    providerOptions?: LanguageModelV2ProviderOptions;
+    providerMetadata?: LanguageModelV1ProviderMetadata;
 }
 /**
 Tool result content part of a prompt. It contains the result of the tool call with the matching ID.
  */
-interface LanguageModelV2ToolResultPart {
+interface LanguageModelV1ToolResultPart {
     type: 'tool-result';
     /**
   ID of the tool call that this result is associated with.
@@ -1485,17 +1525,17 @@ Mime type of the image.
         mimeType?: string;
     }>;
     /**
-     * Additional provider-specific options. They are passed through
+     * Additional provider-specific metadata. They are passed through
      * to the provider from the AI SDK and enable provider-specific
      * functionality that can be fully encapsulated in the provider.
      */
-    providerOptions?: LanguageModelV2ProviderOptions;
+    providerMetadata?: LanguageModelV1ProviderMetadata;
 }
 
 /**
 The configuration of a tool that is defined by the provider.
  */
-type LanguageModelV2ProviderDefinedTool = {
+type LanguageModelV1ProviderDefinedTool = {
     /**
   The type of the tool (always 'provider-defined').
      */
@@ -1514,7 +1554,7 @@ type LanguageModelV2ProviderDefinedTool = {
     args: Record<string, unknown>;
 };
 
-type LanguageModelV2ToolChoice = {
+type LanguageModelV1ToolChoice = {
     type: 'auto';
 } | {
     type: 'none';
@@ -1525,7 +1565,7 @@ type LanguageModelV2ToolChoice = {
     toolName: string;
 };
 
-type LanguageModelV2CallOptions = {
+type LanguageModelV1CallOptions = LanguageModelV1CallSettings & {
     /**
   Whether the user provided the input as messages or as
   a prompt. This can help guide non-chat models in the
@@ -1534,62 +1574,29 @@ type LanguageModelV2CallOptions = {
      */
     inputFormat: 'messages' | 'prompt';
     /**
-  A language mode prompt is a standardized prompt type.
+  The mode affects the behavior of the language model. It is required to
+  support provider-independent streaming and generation of structured objects.
+  The model can take this information and e.g. configure json mode, the correct
+  low level grammar, etc. It can also be used to optimize the efficiency of the
+  streaming, e.g. tool-delta stream parts are only needed in the
+  object-tool mode.
   
-  Note: This is **not** the user-facing prompt. The AI SDK methods will map the
-  user-facing prompt types such as chat or instruction prompts to this format.
-  That approach allows us to evolve the user  facing prompts without breaking
-  the language model interface.
-     */
-    prompt: LanguageModelV2Prompt;
-    /**
-  Maximum number of tokens to generate.
+  @deprecated mode will be removed in v2.
+  All necessary settings will be directly supported through the call settings,
+  in particular responseFormat, toolChoice, and tools.
      */
-    maxTokens?: number;
-    /**
-  Temperature setting.
-  
-  It is recommended to set either `temperature` or `topP`, but not both.
-   */
-    temperature?: number;
-    /**
-  Stop sequences.
-  If set, the model will stop generating text when one of the stop sequences is generated.
-  Providers may have limits on the number of stop sequences.
-   */
-    stopSequences?: string[];
-    /**
-  Nucleus sampling.
-  
-  It is recommended to set either `temperature` or `topP`, but not both.
-   */
-    topP?: number;
-    /**
-  Only sample from the top K options for each subsequent token.
-  
-  Used to remove "long tail" low probability responses.
-  Recommended for advanced use cases only. You usually only need to use temperature.
-   */
-    topK?: number;
-    /**
-  Presence penalty setting. It affects the likelihood of the model to
-  repeat information that is already in the prompt.
-   */
-    presencePenalty?: number;
-    /**
-  Frequency penalty setting. It affects the likelihood of the model
-  to repeatedly use the same words or phrases.
-   */
-    frequencyPenalty?: number;
-    /**
-  Response format. The output can either be text or JSON. Default is text.
-  
-  If JSON is selected, a schema can optionally be provided to guide the LLM.
-   */
-    responseFormat?: {
-        type: 'text';
+    mode: {
+        type: 'regular';
+        /**
+The tools that are available for the model.
+         */
+        tools?: Array<LanguageModelV1FunctionTool | LanguageModelV1ProviderDefinedTool>;
+        /**
+Specifies how the tool should be selected. Defaults to 'auto'.
+         */
+        toolChoice?: LanguageModelV1ToolChoice;
     } | {
-        type: 'json';
+        type: 'object-json';
         /**
          * JSON schema that the generated output should conform to.
          */
@@ -1602,48 +1609,38 @@ type LanguageModelV2CallOptions = {
          * Description of the output that should be generated. Used by some providers for additional LLM guidance.
          */
         description?: string;
+    } | {
+        type: 'object-tool';
+        tool: LanguageModelV1FunctionTool;
     };
     /**
-  The seed (integer) to use for random sampling. If set and supported
-  by the model, calls will generate deterministic results.
-   */
-    seed?: number;
-    /**
-  The tools that are available for the model.
-    */
-    tools?: Array<LanguageModelV2FunctionTool | LanguageModelV2ProviderDefinedTool>;
-    /**
-  Specifies how the tool should be selected. Defaults to 'auto'.
-  */
-    toolChoice?: LanguageModelV2ToolChoice;
-    /**
-  Abort signal for cancelling the operation.
-   */
-    abortSignal?: AbortSignal;
-    /**
-  Additional HTTP headers to be sent with the request.
-  Only applicable for HTTP-based providers.
-   */
-    headers?: Record<string, string | undefined>;
+  A language mode prompt is a standardized prompt type.
+  
+  Note: This is **not** the user-facing prompt. The AI SDK methods will map the
+  user-facing prompt types such as chat or instruction prompts to this format.
+  That approach allows us to evolve the user  facing prompts without breaking
+  the language model interface.
+     */
+    prompt: LanguageModelV1Prompt;
     /**
-     * Additional provider-specific options. They are passed through
-     * to the provider from the AI SDK and enable provider-specific
-     * functionality that can be fully encapsulated in the provider.
+  Additional provider-specific metadata.
+  The metadata is passed through to the provider from the AI SDK and enables
+  provider-specific functionality that can be fully encapsulated in the provider.
      */
-    providerOptions?: LanguageModelV2ProviderOptions;
+    providerMetadata?: LanguageModelV1ProviderMetadata;
 };
 
 /**
 Warning from the model provider for this call. The call will proceed, but e.g.
 some settings might not be supported, which can lead to suboptimal results.
  */
-type LanguageModelV2CallWarning = {
+type LanguageModelV1CallWarning = {
     type: 'unsupported-setting';
-    setting: Omit<keyof LanguageModelV2CallOptions, 'prompt'>;
+    setting: keyof LanguageModelV1CallSettings;
     details?: string;
 } | {
     type: 'unsupported-tool';
-    tool: LanguageModelV2FunctionTool | LanguageModelV2ProviderDefinedTool;
+    tool: LanguageModelV1FunctionTool | LanguageModelV1ProviderDefinedTool;
     details?: string;
 } | {
     type: 'other';
@@ -1662,9 +1659,9 @@ Can be one of the following:
 - `other`: model stopped for other reasons
 - `unknown`: the model has not transmitted a finish reason
  */
-type LanguageModelV2FinishReason = 'stop' | 'length' | 'content-filter' | 'tool-calls' | 'error' | 'other' | 'unknown';
+type LanguageModelV1FinishReason = 'stop' | 'length' | 'content-filter' | 'tool-calls' | 'error' | 'other' | 'unknown';
 
-type LanguageModelV2FunctionToolCall = {
+type LanguageModelV1FunctionToolCall = {
     toolCallType: 'function';
     toolCallId: string;
     toolName: string;
@@ -1678,7 +1675,7 @@ type LanguageModelV2FunctionToolCall = {
 /**
 Log probabilities for each token and its top log probabilities.
  */
-type LanguageModelV2LogProbs = Array<{
+type LanguageModelV1LogProbs = Array<{
     token: string;
     logprob: number;
     topLogprobs: Array<{
@@ -1688,58 +1685,9 @@ type LanguageModelV2LogProbs = Array<{
 }>;
 
 /**
- * Additional provider-specific metadata.
- * Metadata are additional outputs from the provider.
- * They are passed through to the provider from the AI SDK
- * and enable provider-specific functionality
- * that can be fully encapsulated in the provider.
- *
- * This enables us to quickly ship provider-specific functionality
- * without affecting the core AI SDK.
- *
- * The outer record is keyed by the provider name, and the inner
- * record is keyed by the provider-specific metadata key.
- *
- * ```ts
- * {
- *   "anthropic": {
- *     "cacheControl": { "type": "ephemeral" }
- *   }
- * }
- * ```
- */
-type LanguageModelV2ProviderMetadata = Record<string, Record<string, JSONValue>>;
-
-/**
- * A source that has been used as input to generate the response.
- */
-type LanguageModelV2Source = {
-    /**
-     * A URL source. This is return by web search RAG models.
-     */
-    sourceType: 'url';
-    /**
-     * The ID of the source.
-     */
-    id: string;
-    /**
-     * The URL of the source.
-     */
-    url: string;
-    /**
-     * The title of the source.
-     */
-    title?: string;
-    /**
-     * Additional provider metadata for the source.
-     */
-    providerMetadata?: LanguageModelV2ProviderMetadata;
-};
-
-/**
-Specification for a language model that implements the language model interface version 2.
+Specification for a language model that implements the language model interface version 1.
  */
-type LanguageModelV2 = {
+type LanguageModelV1 = {
     /**
   The language model must specify which language model interface
   version it implements. This will allow us to evolve the language
@@ -1747,7 +1695,7 @@ type LanguageModelV2 = {
   implementation versions can be handled as a discriminated union
   on our side.
      */
-    readonly specificationVersion: 'v2';
+    readonly specificationVersion: 'v1';
     /**
   Name of the provider for logging purposes.
      */
@@ -1764,7 +1712,7 @@ type LanguageModelV2 = {
   This is needed to generate the best objects possible w/o requiring the
   user to explicitly specify the object generation mode.
      */
-    readonly defaultObjectGenerationMode: LanguageModelV2ObjectGenerationMode;
+    readonly defaultObjectGenerationMode: LanguageModelV1ObjectGenerationMode;
     /**
   Flag whether this model supports image URLs. Default is `true`.
   
@@ -1803,7 +1751,7 @@ type LanguageModelV2 = {
   Naming: "do" prefix to prevent accidental direct usage of the method
   by the user.
      */
-    doGenerate(options: LanguageModelV2CallOptions): PromiseLike<{
+    doGenerate(options: LanguageModelV1CallOptions): PromiseLike<{
         /**
     Text that the model has generated.
     Can be undefined if the model did not generate any text.
@@ -1839,11 +1787,11 @@ An optional signature for verifying that the reasoning originated from the model
     Tool calls that the model has generated.
     Can be undefined if the model did not generate any tool calls.
          */
-        toolCalls?: Array<LanguageModelV2FunctionToolCall>;
+        toolCalls?: Array<LanguageModelV1FunctionToolCall>;
         /**
     Finish reason.
          */
-        finishReason: LanguageModelV2FinishReason;
+        finishReason: LanguageModelV1FinishReason;
         /**
       Usage information.
          */
@@ -1906,24 +1854,24 @@ An optional signature for verifying that the reasoning originated from the model
            */
             modelId?: string;
         };
-        warnings?: LanguageModelV2CallWarning[];
+        warnings?: LanguageModelV1CallWarning[];
         /**
     Additional provider-specific metadata. They are passed through
     from the provider to the AI SDK and enable provider-specific
     results that can be fully encapsulated in the provider.
          */
-        providerMetadata?: LanguageModelV2ProviderMetadata;
+        providerMetadata?: LanguageModelV1ProviderMetadata;
         /**
     Sources that have been used as input to generate the response.
          */
-        sources?: LanguageModelV2Source[];
+        sources?: LanguageModelV1Source[];
         /**
     Logprobs for the completion.
     `undefined` if the mode does not support logprobs or if was not enabled
     
     @deprecated will be changed into a provider-specific extension in v2
          */
-        logprobs?: LanguageModelV2LogProbs;
+        logprobs?: LanguageModelV1LogProbs;
     }>;
     /**
   Generates a language model output (streaming).
@@ -1933,8 +1881,8 @@ An optional signature for verifying that the reasoning originated from the model
      *
   @return A stream of higher-level language model output parts.
      */
-    doStream(options: LanguageModelV2CallOptions): PromiseLike<{
-        stream: ReadableStream<LanguageModelV2StreamPart>;
+    doStream(options: LanguageModelV1CallOptions): PromiseLike<{
+        stream: ReadableStream<LanguageModelV1StreamPart>;
         /**
     Raw prompt and setting information for observability provider integration.
          */
@@ -1972,10 +1920,10 @@ An optional signature for verifying that the reasoning originated from the model
         /**
     Warnings for the call, e.g. unsupported settings.
          */
-        warnings?: Array<LanguageModelV2CallWarning>;
+        warnings?: Array<LanguageModelV1CallWarning>;
     }>;
 };
-type LanguageModelV2StreamPart = {
+type LanguageModelV1StreamPart = {
     type: 'text-delta';
     textDelta: string;
 } | {
@@ -1989,7 +1937,7 @@ type LanguageModelV2StreamPart = {
     data: string;
 } | {
     type: 'source';
-    source: LanguageModelV2Source;
+    source: LanguageModelV1Source;
 } | {
     type: 'file';
     mimeType: string;
@@ -2003,7 +1951,7 @@ be returned as binary data.
     data: string | Uint8Array;
 } | ({
     type: 'tool-call';
-} & LanguageModelV2FunctionToolCall) | {
+} & LanguageModelV1FunctionToolCall) | {
     type: 'tool-call-delta';
     toolCallType: 'function';
     toolCallId: string;
@@ -2016,13 +1964,13 @@ be returned as binary data.
     modelId?: string;
 } | {
     type: 'finish';
-    finishReason: LanguageModelV2FinishReason;
-    providerMetadata?: LanguageModelV2ProviderMetadata;
+    finishReason: LanguageModelV1FinishReason;
+    providerMetadata?: LanguageModelV1ProviderMetadata;
     usage: {
         promptTokens: number;
         completionTokens: number;
     };
-    logprobs?: LanguageModelV2LogProbs;
+    logprobs?: LanguageModelV1LogProbs;
 } | {
     type: 'error';
     error: unknown;
@@ -2031,67 +1979,294 @@ be returned as binary data.
 The object generation modes available for use with a model. `undefined`
 represents no support for object generation.
    */
-type LanguageModelV2ObjectGenerationMode = 'json' | 'tool' | undefined;
+type LanguageModelV1ObjectGenerationMode = 'json' | 'tool' | undefined;
+
+type TranscriptionModelV1ProviderOptions = Record<string, Record<string, JSONValue>>;
+type TranscriptionModelV1CallOptions = {
+    /**
+  Audio data to transcribe.
+  Accepts a `Uint8Array` or `string`, where `string` is a base64 encoded audio file.
+       */
+    audio: Uint8Array | string;
+    /**
+  The IANA media type of the audio data.
+  
+  @see https://www.iana.org/assignments/media-types/media-types.xhtml
+     */
+    mediaType: string;
+    /**
+  Additional provider-specific options that are passed through to the provider
+  as body parameters.
+  
+  The outer record is keyed by the provider name, and the inner
+  record is keyed by the provider-specific metadata key.
+  ```ts
+  {
+  "openai": {
+  "timestampGranularities": ["word"]
+  }
+  }
+  ```
+   */
+    providerOptions?: TranscriptionModelV1ProviderOptions;
+    /**
+  Abort signal for cancelling the operation.
+   */
+    abortSignal?: AbortSignal;
+    /**
+  Additional HTTP headers to be sent with the request.
+  Only applicable for HTTP-based providers.
+   */
+    headers?: Record<string, string | undefined>;
+};
 
 /**
- * Experimental middleware for LanguageModelV2.
- * This type defines the structure for middleware that can be used to modify
- * the behavior of LanguageModelV2 operations.
+Warning from the model provider for this call. The call will proceed, but e.g.
+some settings might not be supported, which can lead to suboptimal results.
  */
-type LanguageModelV2Middleware = {
+type TranscriptionModelV1CallWarning = {
+    type: 'unsupported-setting';
+    setting: keyof TranscriptionModelV1CallOptions;
+    details?: string;
+} | {
+    type: 'other';
+    message: string;
+};
+
+/**
+Transcription model specification version 1.
+ */
+type TranscriptionModelV1 = {
     /**
-     * Middleware specification version. Use `v2` for the current version.
+  The transcription model must specify which transcription model interface
+  version it implements. This will allow us to evolve the transcription
+  model interface and retain backwards compatibility. The different
+  implementation versions can be handled as a discriminated union
+  on our side.
      */
-    middlewareVersion?: 'v2' | undefined;
+    readonly specificationVersion: 'v1';
     /**
-     * Transforms the parameters before they are passed to the language model.
-     * @param options - Object containing the type of operation and the parameters.
-     * @param options.type - The type of operation ('generate' or 'stream').
-     * @param options.params - The original parameters for the language model call.
-     * @returns A promise that resolves to the transformed parameters.
+  Name of the provider for logging purposes.
      */
-    transformParams?: (options: {
-        type: 'generate' | 'stream';
-        params: LanguageModelV2CallOptions;
-    }) => PromiseLike<LanguageModelV2CallOptions>;
+    readonly provider: string;
     /**
-     * Wraps the generate operation of the language model.
-     * @param options - Object containing the generate function, parameters, and model.
-     * @param options.doGenerate - The original generate function.
-     * @param options.doStream - The original stream function.
-     * @param options.params - The parameters for the generate call. If the
-     * `transformParams` middleware is used, this will be the transformed parameters.
-     * @param options.model - The language model instance.
-     * @returns A promise that resolves to the result of the generate operation.
+  Provider-specific model ID for logging purposes.
      */
-    wrapGenerate?: (options: {
-        doGenerate: () => ReturnType<LanguageModelV2['doGenerate']>;
-        doStream: () => ReturnType<LanguageModelV2['doStream']>;
-        params: LanguageModelV2CallOptions;
-        model: LanguageModelV2;
-    }) => Promise<Awaited<ReturnType<LanguageModelV2['doGenerate']>>>;
+    readonly modelId: string;
     /**
-     * Wraps the stream operation of the language model.
+  Generates a transcript.
+     */
+    doGenerate(options: TranscriptionModelV1CallOptions): PromiseLike<{
+        /**
+         * The complete transcribed text from the audio.
+         */
+        text: string;
+        /**
+         * Array of transcript segments with timing information.
+         * Each segment represents a portion of the transcribed text with start and end times.
+         */
+        segments: Array<{
+            /**
+             * The text content of this segment.
+             */
+            text: string;
+            /**
+             * The start time of this segment in seconds.
+             */
+            startSecond: number;
+            /**
+             * The end time of this segment in seconds.
+             */
+            endSecond: number;
+        }>;
+        /**
+         * The detected language of the audio content, as an ISO-639-1 code (e.g., 'en' for English).
+         * May be undefined if the language couldn't be detected.
+         */
+        language: string | undefined;
+        /**
+         * The total duration of the audio file in seconds.
+         * May be undefined if the duration couldn't be determined.
+         */
+        durationInSeconds: number | undefined;
+        /**
+    Warnings for the call, e.g. unsupported settings.
+         */
+        warnings: Array<TranscriptionModelV1CallWarning>;
+        /**
+    Optional request information for telemetry and debugging purposes.
+         */
+        request?: {
+            /**
+      Raw request HTTP body that was sent to the provider API as a string (JSON should be stringified).
+      Non-HTTP(s) providers should not set this.
+             */
+            body?: string;
+        };
+        /**
+    Response information for telemetry and debugging purposes.
+         */
+        response: {
+            /**
+      Timestamp for the start of the generated response.
+            */
+            timestamp: Date;
+            /**
+      The ID of the response model that was used to generate the response.
+            */
+            modelId: string;
+            /**
+      Response headers.
+            */
+            headers?: SharedV2Headers;
+            /**
+      Response body.
+            */
+            body?: unknown;
+        };
+        /**
+    Additional provider-specific metadata. They are passed through
+    from the provider to the AI SDK and enable provider-specific
+    results that can be fully encapsulated in the provider.
+         */
+        providerMetadata?: Record<string, Record<string, JSONValue>>;
+    }>;
+};
+
+type SpeechModelV1ProviderOptions = Record<string, Record<string, JSONValue>>;
+type SpeechModelV1CallOptions = {
+    /**
+     * Text to convert to speech.
+     */
+    text: string;
+    /**
+     * The voice to use for speech synthesis.
+     * This is provider-specific and may be a voice ID, name, or other identifier.
+     */
+    voice?: string;
+    /**
+     * The desired output format for the audio e.g. "mp3", "wav", etc.
+     */
+    outputFormat?: string;
+    /**
+     * Instructions for the speech generation e.g. "Speak in a slow and steady tone".
+     */
+    instructions?: string;
+    /**
+     * The speed of the speech generation.
+     */
+    speed?: number;
+    /**
+     * Additional provider-specific options that are passed through to the provider
+     * as body parameters.
      *
-     * @param options - Object containing the stream function, parameters, and model.
-     * @param options.doGenerate - The original generate function.
-     * @param options.doStream - The original stream function.
-     * @param options.params - The parameters for the stream call. If the
-     * `transformParams` middleware is used, this will be the transformed parameters.
-     * @param options.model - The language model instance.
-     * @returns A promise that resolves to the result of the stream operation.
+     * The outer record is keyed by the provider name, and the inner
+     * record is keyed by the provider-specific metadata key.
+     * ```ts
+     * {
+     *   "openai": {}
+     * }
+     * ```
      */
-    wrapStream?: (options: {
-        doGenerate: () => ReturnType<LanguageModelV2['doGenerate']>;
-        doStream: () => ReturnType<LanguageModelV2['doStream']>;
-        params: LanguageModelV2CallOptions;
-        model: LanguageModelV2;
-    }) => PromiseLike<Awaited<ReturnType<LanguageModelV2['doStream']>>>;
+    providerOptions?: SpeechModelV1ProviderOptions;
+    /**
+     * Abort signal for cancelling the operation.
+     */
+    abortSignal?: AbortSignal;
+    /**
+     * Additional HTTP headers to be sent with the request.
+     * Only applicable for HTTP-based providers.
+     */
+    headers?: Record<string, string | undefined>;
+};
+
+/**
+ * Warning from the model provider for this call. The call will proceed, but e.g.
+ * some settings might not be supported, which can lead to suboptimal results.
+ */
+type SpeechModelV1CallWarning = {
+    type: 'unsupported-setting';
+    setting: keyof SpeechModelV1CallOptions;
+    details?: string;
+} | {
+    type: 'other';
+    message: string;
 };
+
 /**
- * @deprecated Use `LanguageModelV2Middleware` instead.
+ * Speech model specification version 1.
  */
-type Experimental_LanguageModelV2Middleware = LanguageModelV2Middleware;
+type SpeechModelV1 = {
+    /**
+     * The speech model must specify which speech model interface
+     * version it implements. This will allow us to evolve the speech
+     * model interface and retain backwards compatibility. The different
+     * implementation versions can be handled as a discriminated union
+     * on our side.
+     */
+    readonly specificationVersion: 'v1';
+    /**
+     * Name of the provider for logging purposes.
+     */
+    readonly provider: string;
+    /**
+     * Provider-specific model ID for logging purposes.
+     */
+    readonly modelId: string;
+    /**
+     * Generates speech audio from text.
+     */
+    doGenerate(options: SpeechModelV1CallOptions): PromiseLike<{
+        /**
+         * Generated audio as an ArrayBuffer.
+         * The audio should be returned without any unnecessary conversion.
+         * If the API returns base64 encoded strings, the audio should be returned
+         * as base64 encoded strings. If the API returns binary data, the audio
+         * should be returned as binary data.
+         */
+        audio: string | Uint8Array;
+        /**
+         * Warnings for the call, e.g. unsupported settings.
+         */
+        warnings: Array<SpeechModelV1CallWarning>;
+        /**
+         * Optional request information for telemetry and debugging purposes.
+         */
+        request?: {
+            /**
+             * Response body (available only for providers that use HTTP requests).
+             */
+            body?: unknown;
+        };
+        /**
+         * Response information for telemetry and debugging purposes.
+         */
+        response: {
+            /**
+             * Timestamp for the start of the generated response.
+             */
+            timestamp: Date;
+            /**
+             * The ID of the response model that was used to generate the response.
+             */
+            modelId: string;
+            /**
+             * Response headers.
+             */
+            headers?: SharedV2Headers;
+            /**
+             * Response body.
+             */
+            body?: unknown;
+        };
+        /**
+         * Additional provider-specific metadata. They are passed through
+         * from the provider to the AI SDK and enable provider-specific
+         * results that can be fully encapsulated in the provider.
+         */
+        providerMetadata?: Record<string, Record<string, JSONValue>>;
+    }>;
+};
 
 /**
  * Provider for language, text embedding, and image generation models.
@@ -2118,7 +2293,7 @@ interface ProviderV1 {
   
   @throws {NoSuchModelError} If no such model exists.
      */
-    textEmbeddingModel(modelId: string): EmbeddingModelV1<string>;
+    textEmbeddingModel(modelId: string): EmbeddingModelV2<string>;
     /**
   Returns the image model with the given id.
   The model id is then passed to the provider function to get the model.
@@ -2127,7 +2302,25 @@ interface ProviderV1 {
   
   @returns {ImageModel} The image model associated with the id
   */
-    readonly imageModel?: (modelId: string) => ImageModelV1;
+    readonly imageModel?: (modelId: string) => ImageModelV2;
+    /**
+  Returns the transcription model with the given id.
+  The model id is then passed to the provider function to get the model.
+  
+  @param {string} modelId - The id of the model to return.
+  
+  @returns {TranscriptionModel} The transcription model associated with the id
+  */
+    readonly transcriptionModel?: (modelId: string) => TranscriptionModelV1;
+    /**
+  Returns the speech model with the given id.
+  The model id is then passed to the provider function to get the model.
+  
+  @param {string} modelId - The id of the model to return.
+  
+  @returns {SpeechModel} The speech model associated with the id
+  */
+    readonly speechModel?: (modelId: string) => SpeechModelV1;
 }
 
 /**
@@ -2155,7 +2348,7 @@ interface ProviderV2 {
   
   @throws {NoSuchModelError} If no such model exists.
      */
-    textEmbeddingModel(modelId: string): EmbeddingModelV1<string>;
+    textEmbeddingModel(modelId: string): EmbeddingModelV2<string>;
     /**
   Returns the image model with the given id.
   The model id is then passed to the provider function to get the model.
@@ -2164,7 +2357,7 @@ interface ProviderV2 {
   
   @returns {ImageModel} The image model associated with the id
   */
-    readonly imageModel: (modelId: string) => ImageModelV1;
+    readonly imageModel: (modelId: string) => ImageModelV2;
 }
 
-export { AISDKError, APICallError, type EmbeddingModelV1, type EmbeddingModelV1Embedding, EmptyResponseBodyError, type Experimental_LanguageModelV2Middleware, type ImageModelV1, type ImageModelV1CallOptions, type ImageModelV1CallWarning, InvalidArgumentError, InvalidPromptError, InvalidResponseDataError, type JSONArray, type JSONObject, JSONParseError, type JSONValue, type LanguageModelV1, type LanguageModelV1CallOptions, type LanguageModelV1CallWarning, type LanguageModelV1FilePart, type LanguageModelV1FinishReason, type LanguageModelV1FunctionTool, type LanguageModelV1FunctionToolCall, type LanguageModelV1ImagePart, type LanguageModelV1LogProbs, type LanguageModelV1Message, type LanguageModelV1ObjectGenerationMode, type LanguageModelV1Prompt, type LanguageModelV1ProviderDefinedTool, type LanguageModelV1ProviderMetadata, type LanguageModelV1ReasoningPart, type LanguageModelV1RedactedReasoningPart, type LanguageModelV1Source, type LanguageModelV1StreamPart, type LanguageModelV1TextPart, type LanguageModelV1ToolCallPart, type LanguageModelV1ToolChoice, type LanguageModelV1ToolResultPart, type LanguageModelV2, type LanguageModelV2CallOptions, type LanguageModelV2CallWarning, type LanguageModelV2FilePart, type LanguageModelV2FinishReason, type LanguageModelV2FunctionTool, type LanguageModelV2FunctionToolCall, type LanguageModelV2ImagePart, type LanguageModelV2LogProbs, type LanguageModelV2Message, type LanguageModelV2Middleware, type LanguageModelV2ObjectGenerationMode, type LanguageModelV2Prompt, type LanguageModelV2ProviderDefinedTool, type LanguageModelV2ProviderMetadata, type LanguageModelV2ProviderOptions, type LanguageModelV2ReasoningPart, type LanguageModelV2RedactedReasoningPart, type LanguageModelV2Source, type LanguageModelV2StreamPart, type LanguageModelV2TextPart, type LanguageModelV2ToolCallPart, type LanguageModelV2ToolChoice, type LanguageModelV2ToolResultPart, LoadAPIKeyError, LoadSettingError, NoContentGeneratedError, NoSuchModelError, type ProviderV1, type ProviderV2, TooManyEmbeddingValuesForCallError, TypeValidationError, UnsupportedFunctionalityError, getErrorMessage, isJSONArray, isJSONObject, isJSONValue };
+export { AISDKError, APICallError, type EmbeddingModelV2, type EmbeddingModelV2Embedding, EmptyResponseBodyError, type ImageModelV2, type ImageModelV2CallOptions, type ImageModelV2CallWarning, InvalidArgumentError, InvalidPromptError, InvalidResponseDataError, type JSONArray, type JSONObject, JSONParseError, type JSONValue, type LanguageModelV1, type LanguageModelV1CallOptions, type LanguageModelV1CallWarning, type LanguageModelV1FilePart, type LanguageModelV1FinishReason, type LanguageModelV1FunctionTool, type LanguageModelV1FunctionToolCall, type LanguageModelV1ImagePart, type LanguageModelV1LogProbs, type LanguageModelV1Message, type LanguageModelV1ObjectGenerationMode, type LanguageModelV1Prompt, type LanguageModelV1ProviderDefinedTool, type LanguageModelV1ProviderMetadata, type LanguageModelV1ReasoningPart, type LanguageModelV1RedactedReasoningPart, type LanguageModelV1Source, type LanguageModelV1StreamPart, type LanguageModelV1TextPart, type LanguageModelV1ToolCallPart, type LanguageModelV1ToolChoice, type LanguageModelV1ToolResultPart, type LanguageModelV2, type LanguageModelV2CallOptions, type LanguageModelV2CallWarning, type LanguageModelV2Content, type LanguageModelV2DataContent, type LanguageModelV2File, type LanguageModelV2FilePart, type LanguageModelV2FinishReason, type LanguageModelV2FunctionTool, type LanguageModelV2Message, type LanguageModelV2Middleware, type LanguageModelV2Prompt, type LanguageModelV2ProviderDefinedTool, type LanguageModelV2Reasoning, type LanguageModelV2ReasoningPart, type LanguageModelV2ResponseMetadata, type LanguageModelV2Source, type LanguageModelV2StreamPart, type LanguageModelV2Text, type LanguageModelV2TextPart, type LanguageModelV2ToolCall, type LanguageModelV2ToolCallDelta, type LanguageModelV2ToolCallPart, type LanguageModelV2ToolChoice, type LanguageModelV2ToolResultPart, type LanguageModelV2Usage, LoadAPIKeyError, LoadSettingError, NoContentGeneratedError, NoSuchModelError, type ProviderV1, type ProviderV2, type SharedV2Headers, type SharedV2ProviderMetadata, type SharedV2ProviderOptions, type SpeechModelV1, type SpeechModelV1CallOptions, type SpeechModelV1CallWarning, TooManyEmbeddingValuesForCallError, type TranscriptionModelV1, type TranscriptionModelV1CallOptions, type TranscriptionModelV1CallWarning, TypeValidationError, UnsupportedFunctionalityError, getErrorMessage, isJSONArray, isJSONObject, isJSONValue };