sarvam-ai-sdk 0.2.0-beta → 0.3.0-beta

package/README.md

@@ -1,6 +1,10 @@
+<a href="https://github.com/rajatsandeepsen/sarvam-ai-sdk">
+    <img alt="cover" src="https://github.com/rajatsandeepsen/sarvam-ai-sdk/blob/master/cover.png?raw=true" />
+</a>
+
 # AI SDK - Sarvam Provider
 
-The **[Sarvam provider](https://v5.ai-sdk.dev/providers/ai-sdk-providers/sarvam)** for the [AI SDK](https://v5.ai-sdk.dev/docs)
+The **[Sarvam provider](https://v6.ai-sdk.dev/providers/community-providers/sarvam)** for the [AI SDK](https://v6.ai-sdk.dev/docs)
 contains language model support for the Sarvam chat completion, Text-to-Speech and Speech-to-Text APIs.
 
 ## Setup
@@ -8,11 +12,11 @@ contains language model support for the Sarvam chat completion, Text-to-Speech a
 The **[Sarvam](http://sarvam.ai)** provider is available in the `sarvam-ai-sdk` module. You can install it with
 
 ```bash
-npm i sarvam-ai-sdk ai@5
+npm i sarvam-ai-sdk ai@6
 ```
 
 > [!WARNING]
-> This package only works with Vercel AI-SDK v5, not v6 or latest v7. Make sure to install `ai@5` in your project.
+> This package only works with Vercel AI-SDK v6, not v7. Make sure to install `ai@6` in your project.
 
 ## Provider Instance
 
@@ -238,4 +242,4 @@ sarvam.speechTranslation("saaras:v3");
 
 ## Documentation
 
-Please check out the **[Sarvam provider documentation](https://v5.ai-sdk.dev/providers/ai-sdk-providers/sarvam)** and **[Sarvam API documentation](https://docs.sarvam.ai)** for more information.
+Please check out the **[Sarvam provider documentation](https://v6.ai-sdk.dev/providers/community-providers/sarvam)** and **[Sarvam API documentation](https://docs.sarvam.ai)** for more information.

package/dist/index.d.mts

@@ -1,5 +1,5 @@
 import { FetchFunction } from "@ai-sdk/provider-utils";
-import { LanguageModelV2, SpeechModelV2, TranscriptionModelV2 } from "@ai-sdk/provider";
+import { LanguageModelV3, SpeechModelV3, TranscriptionModelV3 } from "@ai-sdk/provider";
 import z$1, { z } from "zod";
 
 //#region src/config.d.ts
@@ -471,7 +471,7 @@ type SarvamProvider = {
    * 		prompt: "Translate this to malayalam: 'Keep cooking, guys'",
    * 	});
    */
-  (modelId: ChatModelId, settings?: ChatSettings): LanguageModelV2;
+  (modelId: ChatModelId, settings?: ChatSettings): LanguageModelV3;
   /**
    * Creates an Sarvam chat model for text generation.
    *
@@ -481,7 +481,7 @@ type SarvamProvider = {
    * 		prompt: "Translate this to malayalam: 'Keep cooking, guys'",
    * 	});
    */
-  languageModel(modelId: ChatModelId, settings?: ChatSettings): LanguageModelV2;
+  languageModel(modelId: ChatModelId, settings?: ChatSettings): LanguageModelV3;
   /**
    * Creates a Sarvam model for chat.
    *
@@ -491,7 +491,7 @@ type SarvamProvider = {
    * 		prompt: "Translate this to malayalam: 'Keep cooking, guys'",
    * 	});
    */
-  chat(modelId: ChatModelId, settings?: ChatSettings): LanguageModelV2;
+  chat(modelId: ChatModelId, settings?: ChatSettings): LanguageModelV3;
   /**
    * Creates a Sarvam model for transcription.
    *
@@ -508,7 +508,7 @@ type SarvamProvider = {
    *
    * @default unknown
    */
-  languageCode?: (T extends "saaras:v3" ? MoreSarvamLanguageCode : never) | SarvamLanguageCode | "unknown", settings?: TranscriptionSettings<T>): TranscriptionModelV2;
+  languageCode?: (T extends "saaras:v3" ? MoreSarvamLanguageCode : never) | SarvamLanguageCode | "unknown", settings?: TranscriptionSettings<T>): TranscriptionModelV3;
   /**
    * Creates a Sarvam model for Speech translation.
    *
@@ -518,7 +518,7 @@ type SarvamProvider = {
    *		audio: await readFile("./audio.wav"),
    * 	});
    */
-  speechTranslation<T extends SpeechTranslationModelId>(modelId: T, settings?: SpeechTranslationSettings): TranscriptionModelV2;
+  speechTranslation<T extends SpeechTranslationModelId>(modelId: T, settings?: SpeechTranslationSettings): TranscriptionModelV3;
   /**
    * Creates a Sarvam model for speech.
    * @example
@@ -529,7 +529,7 @@ type SarvamProvider = {
    *
    * 	await writeFile("./audio.wav", Buffer.from(audio.base64, "base64"););
    */
-  speech<T extends SpeechModelId>(modelId: T, languageCode: SarvamLanguageCode, settings?: SpeechSettings<T>): SpeechModelV2;
+  speech<T extends SpeechModelId>(modelId: T, languageCode: SarvamLanguageCode, settings?: SpeechSettings<T>): SpeechModelV3;
   /**
    * Creates an Sarvam model for transliterate.
    *
@@ -542,7 +542,7 @@ type SarvamProvider = {
    *		prompt: "eda mone, happy alle?",
    *	});
    */
-  transliterate<T extends SarvamLanguageCode>(settings: TransliterateSettings<false, T>): LanguageModelV2;
+  transliterate<T extends SarvamLanguageCode>(settings: TransliterateSettings<false, T>): LanguageModelV3;
   /**
    * Creates an Sarvam model for translation.
    *
@@ -555,7 +555,7 @@ type SarvamProvider = {
    *		prompt: "ഇതൊക്കെ ശ്രദ്ധിക്കണ്ടേ അംബാനെ?",
    *	});
    */
-  translation<T extends TranslationModelId>(model: T, settings: TranslationSettings<T>): LanguageModelV2;
+  translation<T extends TranslationModelId>(model: T, settings: TranslationSettings<T>): LanguageModelV3;
   /**
    * Creates an Sarvam model for language identification.
    *
@@ -565,7 +565,7 @@ type SarvamProvider = {
    *		prompt: "ബുദ്ധിയാണ് സാറേ ഇവൻ്റെ മെയിൻ",
    *	});
    */
-  languageIdentification(): LanguageModelV2;
+  languageIdentification(): LanguageModelV3;
 };
 //#endregion
 //#region src/provider.d.ts

package/dist/index.d.ts

@@ -1,6 +1,6 @@
 import { FetchFunction } from "@ai-sdk/provider-utils";
 import z$1, { z } from "zod";
-import { LanguageModelV2, SpeechModelV2, TranscriptionModelV2 } from "@ai-sdk/provider";
+import { LanguageModelV3, SpeechModelV3, TranscriptionModelV3 } from "@ai-sdk/provider";
 
 //#region src/config.d.ts
 
@@ -471,7 +471,7 @@ type SarvamProvider = {
    * 		prompt: "Translate this to malayalam: 'Keep cooking, guys'",
    * 	});
    */
-  (modelId: ChatModelId, settings?: ChatSettings): LanguageModelV2;
+  (modelId: ChatModelId, settings?: ChatSettings): LanguageModelV3;
   /**
    * Creates an Sarvam chat model for text generation.
    *
@@ -481,7 +481,7 @@ type SarvamProvider = {
    * 		prompt: "Translate this to malayalam: 'Keep cooking, guys'",
    * 	});
    */
-  languageModel(modelId: ChatModelId, settings?: ChatSettings): LanguageModelV2;
+  languageModel(modelId: ChatModelId, settings?: ChatSettings): LanguageModelV3;
   /**
    * Creates a Sarvam model for chat.
    *
@@ -491,7 +491,7 @@ type SarvamProvider = {
    * 		prompt: "Translate this to malayalam: 'Keep cooking, guys'",
    * 	});
    */
-  chat(modelId: ChatModelId, settings?: ChatSettings): LanguageModelV2;
+  chat(modelId: ChatModelId, settings?: ChatSettings): LanguageModelV3;
   /**
    * Creates a Sarvam model for transcription.
    *
@@ -508,7 +508,7 @@ type SarvamProvider = {
    *
    * @default unknown
    */
-  languageCode?: (T extends "saaras:v3" ? MoreSarvamLanguageCode : never) | SarvamLanguageCode | "unknown", settings?: TranscriptionSettings<T>): TranscriptionModelV2;
+  languageCode?: (T extends "saaras:v3" ? MoreSarvamLanguageCode : never) | SarvamLanguageCode | "unknown", settings?: TranscriptionSettings<T>): TranscriptionModelV3;
   /**
    * Creates a Sarvam model for Speech translation.
    *
@@ -518,7 +518,7 @@ type SarvamProvider = {
    *		audio: await readFile("./audio.wav"),
    * 	});
    */
-  speechTranslation<T extends SpeechTranslationModelId>(modelId: T, settings?: SpeechTranslationSettings): TranscriptionModelV2;
+  speechTranslation<T extends SpeechTranslationModelId>(modelId: T, settings?: SpeechTranslationSettings): TranscriptionModelV3;
   /**
    * Creates a Sarvam model for speech.
    * @example
@@ -529,7 +529,7 @@ type SarvamProvider = {
    *
    * 	await writeFile("./audio.wav", Buffer.from(audio.base64, "base64"););
    */
-  speech<T extends SpeechModelId>(modelId: T, languageCode: SarvamLanguageCode, settings?: SpeechSettings<T>): SpeechModelV2;
+  speech<T extends SpeechModelId>(modelId: T, languageCode: SarvamLanguageCode, settings?: SpeechSettings<T>): SpeechModelV3;
   /**
    * Creates an Sarvam model for transliterate.
    *
@@ -542,7 +542,7 @@ type SarvamProvider = {
    *		prompt: "eda mone, happy alle?",
    *	});
    */
-  transliterate<T extends SarvamLanguageCode>(settings: TransliterateSettings<false, T>): LanguageModelV2;
+  transliterate<T extends SarvamLanguageCode>(settings: TransliterateSettings<false, T>): LanguageModelV3;
   /**
    * Creates an Sarvam model for translation.
    *
@@ -555,7 +555,7 @@ type SarvamProvider = {
    *		prompt: "ഇതൊക്കെ ശ്രദ്ധിക്കണ്ടേ അംബാനെ?",
    *	});
    */
-  translation<T extends TranslationModelId>(model: T, settings: TranslationSettings<T>): LanguageModelV2;
+  translation<T extends TranslationModelId>(model: T, settings: TranslationSettings<T>): LanguageModelV3;
   /**
    * Creates an Sarvam model for language identification.
    *
@@ -565,7 +565,7 @@ type SarvamProvider = {
    *		prompt: "ബുദ്ധിയാണ് സാറേ ഇവൻ്റെ മെയിൻ",
    *	});
    */
-  languageIdentification(): LanguageModelV2;
+  languageIdentification(): LanguageModelV3;
 };
 //#endregion
 //#region src/provider.d.ts

package/dist/index.js

@@ -133,7 +133,7 @@ function prepareTools({ tools, toolChoice }) {
 		toolWarnings
 	};
 	const sarvamTools = [];
-	for (const tool of finalTools) if (tool.type === "provider-defined") toolWarnings.push({
+	for (const tool of finalTools) if (tool.type === "provider") toolWarnings.push({
 		type: "unsupported-tool",
 		tool
 	});
@@ -243,14 +243,14 @@ const chatChunkSchema = zod.default.union([zod.default.object({
 
 //#endregion
 //#region src/chat/utils.ts
-function mapSarvamFinishReason(finishReason) {
+function mapFinishReason(finishReason) {
 	switch (finishReason) {
 		case "stop": return "stop";
 		case "length": return "length";
 		case "content_filter": return "content-filter";
 		case "function_call":
 		case "tool_calls": return "tool-calls";
-		default: return "unknown";
+		default: return "other";
 	}
 }
 function getResponseMetadata({ id, model, created }) {
@@ -265,7 +265,7 @@ function getResponseMetadata({ id, model, created }) {
 //#region src/chat/language-model.ts
 var SarvamChatLanguageModel = class {
 	constructor(modelId, settings, config) {
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 		this.modelId = modelId;
 		this.settings = settings;
 		this.config = config;
@@ -280,8 +280,8 @@ var SarvamChatLanguageModel = class {
 		const { prompt, maxOutputTokens, temperature, topP, topK, frequencyPenalty, presencePenalty, stopSequences, responseFormat, seed, tools, toolChoice, providerOptions, stream } = options;
 		const warnings = [];
 		if (topK) warnings.push({
-			type: "unsupported-setting",
-			setting: "topK"
+			type: "unsupported",
+			feature: "topK"
 		});
 		const sarvamOptions = await (0, __ai_sdk_provider_utils.parseProviderOptions)({
 			provider: "sarvam",
@@ -377,11 +377,19 @@ var SarvamChatLanguageModel = class {
 		}
 		return {
 			content,
-			finishReason: mapSarvamFinishReason(choice.finish_reason),
+			finishReason: mapFinishReason(choice.finish_reason),
 			usage: {
-				inputTokens: response.usage?.prompt_tokens ?? 0,
-				outputTokens: response.usage?.completion_tokens ?? 0,
-				totalTokens: (response.usage?.prompt_tokens ?? 0) + (response.usage?.completion_tokens ?? 0)
+				inputTokens: {
+					total: response.usage?.prompt_tokens ?? void 0,
+					noCache: void 0,
+					cacheRead: void 0,
+					cacheWrite: void 0
+				},
+				outputTokens: {
+					total: response.usage?.completion_tokens ?? void 0,
+					text: void 0,
+					reasoning: void 0
+				}
 			},
 			warnings,
 			request: { body },
@@ -392,7 +400,7 @@ var SarvamChatLanguageModel = class {
 		};
 	}
 	async doStream(options) {
-		const { args, warnings } = await this.getArgs({
+		const { args } = await this.getArgs({
 			...options,
 			stream: true
 		});
@@ -416,10 +424,22 @@ var SarvamChatLanguageModel = class {
 			fetch: this.config.fetch
 		});
 		const toolCalls = [];
-		let finishReason = "unknown";
+		let finishReason = {
+			unified: "other",
+			raw: void 0
+		};
 		let usage = {
-			inputTokens: void 0,
-			outputTokens: void 0
+			inputTokens: {
+				total: void 0,
+				noCache: void 0,
+				cacheRead: void 0,
+				cacheWrite: void 0
+			},
+			outputTokens: {
+				total: void 0,
+				text: void 0,
+				reasoning: void 0
+			}
 		};
 		let isFirstChunk = true;
 		this.config;
@@ -427,7 +447,10 @@ var SarvamChatLanguageModel = class {
 			stream: response.pipeThrough(new TransformStream({
 				transform(chunk, controller) {
 					if (!chunk.success) {
-						finishReason = "error";
+						finishReason = {
+							unified: "error",
+							raw: void 0
+						};
 						controller.enqueue({
 							type: "error",
 							error: chunk.error
@@ -436,7 +459,10 @@ var SarvamChatLanguageModel = class {
 					}
 					const value = chunk.value;
 					if ("error" in value) {
-						finishReason = "error";
+						finishReason = {
+							unified: "error",
+							raw: void 0
+						};
 						controller.enqueue({
 							type: "error",
 							error: value.error
@@ -452,11 +478,23 @@ var SarvamChatLanguageModel = class {
 						});
 					}
 					if (value.x_sarvam?.usage != null) usage = {
-						inputTokens: value.x_sarvam.usage.prompt_tokens ?? void 0,
-						outputTokens: value.x_sarvam.usage.completion_tokens ?? void 0
+						inputTokens: {
+							total: value.x_sarvam.usage.prompt_tokens ?? void 0,
+							noCache: void 0,
+							cacheRead: void 0,
+							cacheWrite: void 0
+						},
+						outputTokens: {
+							total: value.x_sarvam.usage.completion_tokens ?? void 0,
+							text: void 0,
+							reasoning: void 0
+						}
 					};
 					const choice = value.choices[0];
-					if (choice?.finish_reason != null) finishReason = mapSarvamFinishReason(choice.finish_reason);
+					if (choice?.finish_reason != null) finishReason = {
+						unified: mapFinishReason(choice.finish_reason),
+						raw: choice.finish_reason
+					};
 					if (choice?.delta == null) return;
 					const delta = choice.delta;
 					if (delta.reasoning != null && delta.reasoning.length > 0) controller.enqueue({
@@ -532,11 +570,7 @@ var SarvamChatLanguageModel = class {
 					controller.enqueue({
 						type: "finish",
 						finishReason,
-						usage: {
-							inputTokens: usage.inputTokens ?? 0,
-							outputTokens: usage.outputTokens ?? 0,
-							totalTokens: (usage.inputTokens ?? 0) + (usage.outputTokens ?? 0)
-						}
+						usage
 					});
 				}
 			})),
@@ -596,7 +630,7 @@ var SarvamSpeechTranslationModel = class {
 	constructor(modelId, config) {
 		this.modelId = modelId;
 		this.config = config;
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 	}
 	get provider() {
 		return this.config.provider;
@@ -606,7 +640,6 @@ var SarvamSpeechTranslationModel = class {
 	}
 	async getArgs(options) {
 		const { audio, mediaType, providerOptions } = options;
-		const warnings = [];
 		const sarvamOptions = await (0, __ai_sdk_provider_utils.parseProviderOptions)({
 			provider: "sarvam",
 			providerOptions: { sarvam: {
@@ -624,7 +657,7 @@ var SarvamSpeechTranslationModel = class {
 		});
 		return {
 			formData,
-			warnings
+			warnings: []
 		};
 	}
 	async doGenerate(options) {
@@ -747,7 +780,7 @@ var SarvamTranscriptionModel = class {
 		this.modelId = modelId;
 		this.languageCode = languageCode;
 		this.config = config;
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 	}
 	get provider() {
 		return this.config.provider;
@@ -757,7 +790,6 @@ var SarvamTranscriptionModel = class {
 	}
 	async getArgs(options) {
 		const { audio, mediaType, providerOptions } = options;
-		const warnings = [];
 		const sarvamOptions = await (0, __ai_sdk_provider_utils.parseProviderOptions)({
 			provider: "sarvam",
 			providerOptions: { sarvam: {
@@ -780,7 +812,7 @@ var SarvamTranscriptionModel = class {
 		});
 		return {
 			formData,
-			warnings
+			warnings: []
 		};
 	}
 	async doGenerate(options) {
@@ -923,11 +955,10 @@ var SarvamSpeechModel = class {
 		this.modelId = modelId;
 		this.languageCode = languageCode;
 		this.config = config;
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 	}
 	async getArgs(options) {
 		const { text, voice, outputFormat = "wav", speed, providerOptions } = options;
-		const warnings = [];
 		const sarvamOptions = await (0, __ai_sdk_provider_utils.parseProviderOptions)({
 			provider: "sarvam",
 			providerOptions: { sarvam: {
@@ -949,7 +980,7 @@ var SarvamSpeechModel = class {
 		});
 		return {
 			requestBody,
-			warnings
+			warnings: []
 		};
 	}
 	async doGenerate(options) {
@@ -1002,7 +1033,7 @@ const convertPromptToInput = (prompt) => {
 //#region src/ttt/lid-model.ts
 var SarvamLidModel = class {
 	constructor(config) {
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 		this.modelId = "unknown";
 		this.config = config;
 	}
@@ -1020,7 +1051,7 @@ var SarvamLidModel = class {
 		};
 	}
 	async doGenerate(options) {
-		const { args, warnings } = this.getArgs({
+		const { args } = this.getArgs({
 			...options,
 			stream: false
 		});
@@ -1043,11 +1074,19 @@ var SarvamLidModel = class {
 			}],
 			finishReason: "stop",
 			usage: {
-				inputTokens: NaN,
-				outputTokens: NaN,
-				totalTokens: NaN
+				inputTokens: {
+					total: void 0,
+					noCache: void 0,
+					cacheRead: void 0,
+					cacheWrite: void 0
+				},
+				outputTokens: {
+					total: void 0,
+					text: void 0,
+					reasoning: void 0
+				}
 			},
-			warnings
+			warnings: []
 		};
 	}
 	async doStream(_options) {
@@ -1089,7 +1128,7 @@ const translationResponseSchema = zod.default.object({
 //#region src/ttt/translation-model.ts
 var SarvamTranslationModel = class {
 	constructor(modelId, settings, config) {
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 		this.modelId = modelId;
 		this.settings = settings;
 		this.config = config;
@@ -1102,7 +1141,6 @@ var SarvamTranslationModel = class {
 	}
 	async getArgs(options) {
 		const { prompt, providerOptions } = options;
-		const warnings = [];
 		const sarvamOptions = await (0, __ai_sdk_provider_utils.parseProviderOptions)({
 			provider: "sarvam",
 			providerOptions: { sarvam: {
@@ -1127,11 +1165,11 @@ var SarvamTranslationModel = class {
 				source_language_code: from,
 				target_language_code: to
 			},
-			warnings
+			warnings: []
 		};
 	}
 	async doGenerate(options) {
-		const { args, warnings } = await this.getArgs({
+		const { args } = await this.getArgs({
 			...options,
 			stream: false
 		});
@@ -1154,11 +1192,19 @@ var SarvamTranslationModel = class {
 			}],
 			finishReason: "stop",
 			usage: {
-				inputTokens: NaN,
-				outputTokens: NaN,
-				totalTokens: NaN
+				inputTokens: {
+					total: void 0,
+					noCache: void 0,
+					cacheRead: void 0,
+					cacheWrite: void 0
+				},
+				outputTokens: {
+					total: void 0,
+					text: void 0,
+					reasoning: void 0
+				}
 			},
-			warnings
+			warnings: []
 		};
 	}
 	async doStream(_options) {
@@ -1185,7 +1231,7 @@ const transliterateResponseSchema = zod.default.object({
 //#region src/ttt/transliterate-model.ts
 var SarvamTransliterateModel = class {
 	constructor(settings, config) {
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 		this.modelId = "unknown";
 		this.settings = settings;
 		this.config = config;
@@ -1198,7 +1244,6 @@ var SarvamTransliterateModel = class {
 	}
 	async getArgs(options) {
 		const { prompt, providerOptions } = options;
-		const warnings = [];
 		const sarvamOptions = await (0, __ai_sdk_provider_utils.parseProviderOptions)({
 			provider: "sarvam",
 			providerOptions: { sarvam: {
@@ -1223,11 +1268,11 @@ var SarvamTransliterateModel = class {
 				target_language_code: to,
 				spoken_form_numerals_language: sarvamOptions.spoken_form ? sarvamOptions.spoken_form_numerals_language ?? "english" : void 0
 			},
-			warnings
+			warnings: []
 		};
 	}
 	async doGenerate(options) {
-		const { args, warnings } = await this.getArgs({
+		const { args } = await this.getArgs({
 			...options,
 			stream: false
 		});
@@ -1250,11 +1295,19 @@ var SarvamTransliterateModel = class {
 			}],
 			finishReason: "stop",
 			usage: {
-				inputTokens: NaN,
-				outputTokens: NaN,
-				totalTokens: NaN
+				inputTokens: {
+					total: void 0,
+					noCache: void 0,
+					cacheRead: void 0,
+					cacheWrite: void 0
+				},
+				outputTokens: {
+					total: void 0,
+					text: void 0,
+					reasoning: void 0
+				}
 			},
-			warnings
+			warnings: []
 		};
 	}
 	async doStream(_options) {

package/dist/index.mjs

@@ -107,7 +107,7 @@ function prepareTools({ tools, toolChoice }) {
 		toolWarnings
 	};
 	const sarvamTools = [];
-	for (const tool of finalTools) if (tool.type === "provider-defined") toolWarnings.push({
+	for (const tool of finalTools) if (tool.type === "provider") toolWarnings.push({
 		type: "unsupported-tool",
 		tool
 	});
@@ -217,14 +217,14 @@ const chatChunkSchema = z$1.union([z$1.object({
 
 //#endregion
 //#region src/chat/utils.ts
-function mapSarvamFinishReason(finishReason) {
+function mapFinishReason(finishReason) {
 	switch (finishReason) {
 		case "stop": return "stop";
 		case "length": return "length";
 		case "content_filter": return "content-filter";
 		case "function_call":
 		case "tool_calls": return "tool-calls";
-		default: return "unknown";
+		default: return "other";
 	}
 }
 function getResponseMetadata({ id, model, created }) {
@@ -239,7 +239,7 @@ function getResponseMetadata({ id, model, created }) {
 //#region src/chat/language-model.ts
 var SarvamChatLanguageModel = class {
 	constructor(modelId, settings, config) {
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 		this.modelId = modelId;
 		this.settings = settings;
 		this.config = config;
@@ -254,8 +254,8 @@ var SarvamChatLanguageModel = class {
 		const { prompt, maxOutputTokens, temperature, topP, topK, frequencyPenalty, presencePenalty, stopSequences, responseFormat, seed, tools, toolChoice, providerOptions, stream } = options;
 		const warnings = [];
 		if (topK) warnings.push({
-			type: "unsupported-setting",
-			setting: "topK"
+			type: "unsupported",
+			feature: "topK"
 		});
 		const sarvamOptions = await parseProviderOptions({
 			provider: "sarvam",
@@ -351,11 +351,19 @@ var SarvamChatLanguageModel = class {
 		}
 		return {
 			content,
-			finishReason: mapSarvamFinishReason(choice.finish_reason),
+			finishReason: mapFinishReason(choice.finish_reason),
 			usage: {
-				inputTokens: response.usage?.prompt_tokens ?? 0,
-				outputTokens: response.usage?.completion_tokens ?? 0,
-				totalTokens: (response.usage?.prompt_tokens ?? 0) + (response.usage?.completion_tokens ?? 0)
+				inputTokens: {
+					total: response.usage?.prompt_tokens ?? void 0,
+					noCache: void 0,
+					cacheRead: void 0,
+					cacheWrite: void 0
+				},
+				outputTokens: {
+					total: response.usage?.completion_tokens ?? void 0,
+					text: void 0,
+					reasoning: void 0
+				}
 			},
 			warnings,
 			request: { body },
@@ -366,7 +374,7 @@ var SarvamChatLanguageModel = class {
 		};
 	}
 	async doStream(options) {
-		const { args, warnings } = await this.getArgs({
+		const { args } = await this.getArgs({
 			...options,
 			stream: true
 		});
@@ -390,10 +398,22 @@ var SarvamChatLanguageModel = class {
 			fetch: this.config.fetch
 		});
 		const toolCalls = [];
-		let finishReason = "unknown";
+		let finishReason = {
+			unified: "other",
+			raw: void 0
+		};
 		let usage = {
-			inputTokens: void 0,
-			outputTokens: void 0
+			inputTokens: {
+				total: void 0,
+				noCache: void 0,
+				cacheRead: void 0,
+				cacheWrite: void 0
+			},
+			outputTokens: {
+				total: void 0,
+				text: void 0,
+				reasoning: void 0
+			}
 		};
 		let isFirstChunk = true;
 		this.config;
@@ -401,7 +421,10 @@ var SarvamChatLanguageModel = class {
 			stream: response.pipeThrough(new TransformStream({
 				transform(chunk, controller) {
 					if (!chunk.success) {
-						finishReason = "error";
+						finishReason = {
+							unified: "error",
+							raw: void 0
+						};
 						controller.enqueue({
 							type: "error",
 							error: chunk.error
@@ -410,7 +433,10 @@ var SarvamChatLanguageModel = class {
 					}
 					const value = chunk.value;
 					if ("error" in value) {
-						finishReason = "error";
+						finishReason = {
+							unified: "error",
+							raw: void 0
+						};
 						controller.enqueue({
 							type: "error",
 							error: value.error
@@ -426,11 +452,23 @@ var SarvamChatLanguageModel = class {
 						});
 					}
 					if (value.x_sarvam?.usage != null) usage = {
-						inputTokens: value.x_sarvam.usage.prompt_tokens ?? void 0,
-						outputTokens: value.x_sarvam.usage.completion_tokens ?? void 0
+						inputTokens: {
+							total: value.x_sarvam.usage.prompt_tokens ?? void 0,
+							noCache: void 0,
+							cacheRead: void 0,
+							cacheWrite: void 0
+						},
+						outputTokens: {
+							total: value.x_sarvam.usage.completion_tokens ?? void 0,
+							text: void 0,
+							reasoning: void 0
+						}
 					};
 					const choice = value.choices[0];
-					if (choice?.finish_reason != null) finishReason = mapSarvamFinishReason(choice.finish_reason);
+					if (choice?.finish_reason != null) finishReason = {
+						unified: mapFinishReason(choice.finish_reason),
+						raw: choice.finish_reason
+					};
 					if (choice?.delta == null) return;
 					const delta = choice.delta;
 					if (delta.reasoning != null && delta.reasoning.length > 0) controller.enqueue({
@@ -506,11 +544,7 @@ var SarvamChatLanguageModel = class {
 					controller.enqueue({
 						type: "finish",
 						finishReason,
-						usage: {
-							inputTokens: usage.inputTokens ?? 0,
-							outputTokens: usage.outputTokens ?? 0,
-							totalTokens: (usage.inputTokens ?? 0) + (usage.outputTokens ?? 0)
-						}
+						usage
 					});
 				}
 			})),
@@ -570,7 +604,7 @@ var SarvamSpeechTranslationModel = class {
 	constructor(modelId, config) {
 		this.modelId = modelId;
 		this.config = config;
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 	}
 	get provider() {
 		return this.config.provider;
@@ -580,7 +614,6 @@ var SarvamSpeechTranslationModel = class {
 	}
 	async getArgs(options) {
 		const { audio, mediaType, providerOptions } = options;
-		const warnings = [];
 		const sarvamOptions = await parseProviderOptions({
 			provider: "sarvam",
 			providerOptions: { sarvam: {
@@ -598,7 +631,7 @@ var SarvamSpeechTranslationModel = class {
 		});
 		return {
 			formData,
-			warnings
+			warnings: []
 		};
 	}
 	async doGenerate(options) {
@@ -721,7 +754,7 @@ var SarvamTranscriptionModel = class {
 		this.modelId = modelId;
 		this.languageCode = languageCode;
 		this.config = config;
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 	}
 	get provider() {
 		return this.config.provider;
@@ -731,7 +764,6 @@ var SarvamTranscriptionModel = class {
 	}
 	async getArgs(options) {
 		const { audio, mediaType, providerOptions } = options;
-		const warnings = [];
 		const sarvamOptions = await parseProviderOptions({
 			provider: "sarvam",
 			providerOptions: { sarvam: {
@@ -754,7 +786,7 @@ var SarvamTranscriptionModel = class {
 		});
 		return {
 			formData,
-			warnings
+			warnings: []
 		};
 	}
 	async doGenerate(options) {
@@ -897,11 +929,10 @@ var SarvamSpeechModel = class {
 		this.modelId = modelId;
 		this.languageCode = languageCode;
 		this.config = config;
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 	}
 	async getArgs(options) {
 		const { text, voice, outputFormat = "wav", speed, providerOptions } = options;
-		const warnings = [];
 		const sarvamOptions = await parseProviderOptions({
 			provider: "sarvam",
 			providerOptions: { sarvam: {
@@ -923,7 +954,7 @@ var SarvamSpeechModel = class {
 		});
 		return {
 			requestBody,
-			warnings
+			warnings: []
 		};
 	}
 	async doGenerate(options) {
@@ -976,7 +1007,7 @@ const convertPromptToInput = (prompt) => {
 //#region src/ttt/lid-model.ts
 var SarvamLidModel = class {
 	constructor(config) {
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 		this.modelId = "unknown";
 		this.config = config;
 	}
@@ -994,7 +1025,7 @@ var SarvamLidModel = class {
 		};
 	}
 	async doGenerate(options) {
-		const { args, warnings } = this.getArgs({
+		const { args } = this.getArgs({
 			...options,
 			stream: false
 		});
@@ -1017,11 +1048,19 @@ var SarvamLidModel = class {
 			}],
 			finishReason: "stop",
 			usage: {
-				inputTokens: NaN,
-				outputTokens: NaN,
-				totalTokens: NaN
+				inputTokens: {
+					total: void 0,
+					noCache: void 0,
+					cacheRead: void 0,
+					cacheWrite: void 0
+				},
+				outputTokens: {
+					total: void 0,
+					text: void 0,
+					reasoning: void 0
+				}
 			},
-			warnings
+			warnings: []
 		};
 	}
 	async doStream(_options) {
@@ -1063,7 +1102,7 @@ const translationResponseSchema = z$1.object({
 //#region src/ttt/translation-model.ts
 var SarvamTranslationModel = class {
 	constructor(modelId, settings, config) {
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 		this.modelId = modelId;
 		this.settings = settings;
 		this.config = config;
@@ -1076,7 +1115,6 @@ var SarvamTranslationModel = class {
 	}
 	async getArgs(options) {
 		const { prompt, providerOptions } = options;
-		const warnings = [];
 		const sarvamOptions = await parseProviderOptions({
 			provider: "sarvam",
 			providerOptions: { sarvam: {
@@ -1101,11 +1139,11 @@ var SarvamTranslationModel = class {
 				source_language_code: from,
 				target_language_code: to
 			},
-			warnings
+			warnings: []
 		};
 	}
 	async doGenerate(options) {
-		const { args, warnings } = await this.getArgs({
+		const { args } = await this.getArgs({
 			...options,
 			stream: false
 		});
@@ -1128,11 +1166,19 @@ var SarvamTranslationModel = class {
 			}],
 			finishReason: "stop",
 			usage: {
-				inputTokens: NaN,
-				outputTokens: NaN,
-				totalTokens: NaN
+				inputTokens: {
+					total: void 0,
+					noCache: void 0,
+					cacheRead: void 0,
+					cacheWrite: void 0
+				},
+				outputTokens: {
+					total: void 0,
+					text: void 0,
+					reasoning: void 0
+				}
 			},
-			warnings
+			warnings: []
 		};
 	}
 	async doStream(_options) {
@@ -1159,7 +1205,7 @@ const transliterateResponseSchema = z$1.object({
 //#region src/ttt/transliterate-model.ts
 var SarvamTransliterateModel = class {
 	constructor(settings, config) {
-		this.specificationVersion = "v2";
+		this.specificationVersion = "v3";
 		this.modelId = "unknown";
 		this.settings = settings;
 		this.config = config;
@@ -1172,7 +1218,6 @@ var SarvamTransliterateModel = class {
 	}
 	async getArgs(options) {
 		const { prompt, providerOptions } = options;
-		const warnings = [];
 		const sarvamOptions = await parseProviderOptions({
 			provider: "sarvam",
 			providerOptions: { sarvam: {
@@ -1197,11 +1242,11 @@ var SarvamTransliterateModel = class {
 				target_language_code: to,
 				spoken_form_numerals_language: sarvamOptions.spoken_form ? sarvamOptions.spoken_form_numerals_language ?? "english" : void 0
 			},
-			warnings
+			warnings: []
 		};
 	}
 	async doGenerate(options) {
-		const { args, warnings } = await this.getArgs({
+		const { args } = await this.getArgs({
 			...options,
 			stream: false
 		});
@@ -1224,11 +1269,19 @@ var SarvamTransliterateModel = class {
 			}],
 			finishReason: "stop",
 			usage: {
-				inputTokens: NaN,
-				outputTokens: NaN,
-				totalTokens: NaN
+				inputTokens: {
+					total: void 0,
+					noCache: void 0,
+					cacheRead: void 0,
+					cacheWrite: void 0
+				},
+				outputTokens: {
+					total: void 0,
+					text: void 0,
+					reasoning: void 0
+				}
 			},
-			warnings
+			warnings: []
 		};
 	}
 	async doStream(_options) {

package/package.json

@@ -1,7 +1,7 @@
 {
 	"name": "sarvam-ai-sdk",
 	"description": "Sarvam provider support for Vercel's AI-SDK",
-	"version": "0.2.0-beta",
+	"version": "0.3.0-beta",
 	"main": "./dist/index.js",
 	"module": "./dist/index.mjs",
 	"types": "./dist/index.d.ts",
@@ -23,19 +23,19 @@
 		}
 	},
 	"dependencies": {
-		"@ai-sdk/provider": "2",
-		"@ai-sdk/provider-utils": "3"
+		"@ai-sdk/provider": "^3.0.0",
+		"@ai-sdk/provider-utils": "^4.0.0"
 	},
 	"devDependencies": {
 		"@biomejs/biome": "2.4.8",
 		"@types/node": "20.17.24",
-		"ai": "5",
+		"ai": "^6.0.0",
 		"tsdown": "^0.15.10",
 		"typescript": "5.6.3",
 		"zod": "^4.1.8"
 	},
 	"peerDependencies": {
-		"ai": "5",
+		"ai": "^6.0.0",
 		"zod": "^3.25.0 || ^4.0.0"
 	},
 	"engines": {
@@ -51,7 +51,7 @@
 	"bugs": {
 		"url": "https://github.com/rajatsandeepsen/sarvam-ai-sdk/issues"
 	},
-	"homepage": "https://v5.ai-sdk.dev/docs",
+	"homepage": "https://v6.ai-sdk.dev/docs",
 	"keywords": [
 		"ai",
 		"ai-sdk",