@llmgateway/ai-sdk-provider 3.5.0 → 3.6.0

package/README.md

@@ -13,12 +13,12 @@ The [LLMGateway](https://llmgateway.io/) provider for the [Vercel AI SDK](https:
 ## Setup
 
 ```bash
-# For pnpm
-pnpm add @llmgateway/ai-sdk-provider
-
 # For npm
 npm install @llmgateway/ai-sdk-provider
 
+# For pnpm
+pnpm add @llmgateway/ai-sdk-provider
+
 # For yarn
 yarn add @llmgateway/ai-sdk-provider
 ```

package/dist/index.d.mts

@@ -1,5 +1,5 @@
-import { LanguageModelV2, LanguageModelV2CallOptions, LanguageModelV2Content, LanguageModelV2FinishReason, LanguageModelV2Usage, LanguageModelV2CallWarning, LanguageModelV2ResponseMetadata, SharedV2Headers, LanguageModelV2StreamPart, ImageModelV3, ImageModelV3CallOptions, SharedV3Warning } from '@ai-sdk/provider';
-export { LanguageModelV2, LanguageModelV2Prompt } from '@ai-sdk/provider';
+import { LanguageModelV3, LanguageModelV3CallOptions, LanguageModelV3GenerateResult, LanguageModelV3StreamResult, ImageModelV3, ImageModelV3CallOptions, SharedV3Warning } from '@ai-sdk/provider';
+export { LanguageModelV3, LanguageModelV3Prompt } from '@ai-sdk/provider';
 import { models, Provider } from '@llmgateway/models';
 
 type LLMGatewayChatModelId = (typeof models)[number]['id'] | `${Provider}/${(typeof models)[number]['id']}` | 'test-model';
@@ -146,45 +146,17 @@ type LLMGatewayChatConfig = {
     fetch?: typeof fetch;
     extraBody?: Record<string, unknown>;
 };
-declare class LLMGatewayChatLanguageModel implements LanguageModelV2 {
-    readonly specificationVersion: "v2";
+declare class LLMGatewayChatLanguageModel implements LanguageModelV3 {
+    readonly specificationVersion: "v3";
     readonly provider = "llmgateway";
-    readonly defaultObjectGenerationMode: "tool";
     readonly modelId: LLMGatewayChatModelId;
     readonly supportedUrls: Record<string, RegExp[]>;
     readonly settings: LLMGatewayChatSettings;
     private readonly config;
     constructor(modelId: LLMGatewayChatModelId, settings: LLMGatewayChatSettings, config: LLMGatewayChatConfig);
     private getArgs;
-    doGenerate(options: LanguageModelV2CallOptions): Promise<{
-        content: Array<LanguageModelV2Content>;
-        finishReason: LanguageModelV2FinishReason;
-        usage: LanguageModelV2Usage;
-        warnings: Array<LanguageModelV2CallWarning>;
-        providerMetadata?: {
-            llmgateway: {
-                usage: LLMGatewayUsageAccounting;
-            };
-        };
-        request?: {
-            body?: unknown;
-        };
-        response?: LanguageModelV2ResponseMetadata & {
-            headers?: SharedV2Headers;
-            body?: unknown;
-        };
-    }>;
-    doStream(options: LanguageModelV2CallOptions): Promise<{
-        stream: ReadableStream<LanguageModelV2StreamPart>;
-        warnings: Array<LanguageModelV2CallWarning>;
-        request?: {
-            body?: unknown;
-        };
-        response?: LanguageModelV2ResponseMetadata & {
-            headers?: SharedV2Headers;
-            body?: unknown;
-        };
-    }>;
+    doGenerate(options: LanguageModelV3CallOptions): Promise<LanguageModelV3GenerateResult>;
+    doStream(options: LanguageModelV3CallOptions): Promise<LanguageModelV3StreamResult>;
 }
 
 type LLMGatewayCompletionConfig = {
@@ -198,18 +170,17 @@ type LLMGatewayCompletionConfig = {
     fetch?: typeof fetch;
     extraBody?: Record<string, unknown>;
 };
-declare class LLMGatewayCompletionLanguageModel implements LanguageModelV2 {
-    readonly specificationVersion: "v2";
+declare class LLMGatewayCompletionLanguageModel implements LanguageModelV3 {
+    readonly specificationVersion: "v3";
     readonly provider = "llmgateway";
     readonly modelId: LLMGatewayChatModelId;
     readonly supportedUrls: Record<string, RegExp[]>;
-    readonly defaultObjectGenerationMode: undefined;
     readonly settings: LLMGatewayCompletionSettings;
     private readonly config;
     constructor(modelId: LLMGatewayChatModelId, settings: LLMGatewayCompletionSettings, config: LLMGatewayCompletionConfig);
     private getArgs;
-    doGenerate(options: LanguageModelV2CallOptions): Promise<Awaited<ReturnType<LanguageModelV2['doGenerate']>>>;
-    doStream(options: LanguageModelV2CallOptions): Promise<Awaited<ReturnType<LanguageModelV2['doStream']>>>;
+    doGenerate(options: LanguageModelV3CallOptions): Promise<Awaited<ReturnType<LanguageModelV3['doGenerate']>>>;
+    doStream(options: LanguageModelV3CallOptions): Promise<Awaited<ReturnType<LanguageModelV3['doStream']>>>;
 }
 
 type LLMGatewayImageConfig = {
@@ -240,7 +211,7 @@ declare class LLMGatewayImageModel implements ImageModelV3 {
     }>;
 }
 
-interface LLMGatewayProvider extends LanguageModelV2 {
+interface LLMGatewayProvider {
     (modelId: LLMGatewayChatModelId, settings?: LLMGatewayCompletionSettings): LLMGatewayCompletionLanguageModel;
     (modelId: LLMGatewayChatModelId, settings?: LLMGatewayChatSettings): LLMGatewayChatLanguageModel;
     languageModel(modelId: LLMGatewayChatModelId, settings?: LLMGatewayCompletionSettings): LLMGatewayCompletionLanguageModel;

package/dist/index.d.ts

@@ -1,5 +1,5 @@
-import { LanguageModelV2, LanguageModelV2CallOptions, LanguageModelV2Content, LanguageModelV2FinishReason, LanguageModelV2Usage, LanguageModelV2CallWarning, LanguageModelV2ResponseMetadata, SharedV2Headers, LanguageModelV2StreamPart, ImageModelV3, ImageModelV3CallOptions, SharedV3Warning } from '@ai-sdk/provider';
-export { LanguageModelV2, LanguageModelV2Prompt } from '@ai-sdk/provider';
+import { LanguageModelV3, LanguageModelV3CallOptions, LanguageModelV3GenerateResult, LanguageModelV3StreamResult, ImageModelV3, ImageModelV3CallOptions, SharedV3Warning } from '@ai-sdk/provider';
+export { LanguageModelV3, LanguageModelV3Prompt } from '@ai-sdk/provider';
 import { models, Provider } from '@llmgateway/models';
 
 type LLMGatewayChatModelId = (typeof models)[number]['id'] | `${Provider}/${(typeof models)[number]['id']}` | 'test-model';
@@ -146,45 +146,17 @@ type LLMGatewayChatConfig = {
     fetch?: typeof fetch;
     extraBody?: Record<string, unknown>;
 };
-declare class LLMGatewayChatLanguageModel implements LanguageModelV2 {
-    readonly specificationVersion: "v2";
+declare class LLMGatewayChatLanguageModel implements LanguageModelV3 {
+    readonly specificationVersion: "v3";
     readonly provider = "llmgateway";
-    readonly defaultObjectGenerationMode: "tool";
     readonly modelId: LLMGatewayChatModelId;
     readonly supportedUrls: Record<string, RegExp[]>;
     readonly settings: LLMGatewayChatSettings;
     private readonly config;
     constructor(modelId: LLMGatewayChatModelId, settings: LLMGatewayChatSettings, config: LLMGatewayChatConfig);
     private getArgs;
-    doGenerate(options: LanguageModelV2CallOptions): Promise<{
-        content: Array<LanguageModelV2Content>;
-        finishReason: LanguageModelV2FinishReason;
-        usage: LanguageModelV2Usage;
-        warnings: Array<LanguageModelV2CallWarning>;
-        providerMetadata?: {
-            llmgateway: {
-                usage: LLMGatewayUsageAccounting;
-            };
-        };
-        request?: {
-            body?: unknown;
-        };
-        response?: LanguageModelV2ResponseMetadata & {
-            headers?: SharedV2Headers;
-            body?: unknown;
-        };
-    }>;
-    doStream(options: LanguageModelV2CallOptions): Promise<{
-        stream: ReadableStream<LanguageModelV2StreamPart>;
-        warnings: Array<LanguageModelV2CallWarning>;
-        request?: {
-            body?: unknown;
-        };
-        response?: LanguageModelV2ResponseMetadata & {
-            headers?: SharedV2Headers;
-            body?: unknown;
-        };
-    }>;
+    doGenerate(options: LanguageModelV3CallOptions): Promise<LanguageModelV3GenerateResult>;
+    doStream(options: LanguageModelV3CallOptions): Promise<LanguageModelV3StreamResult>;
 }
 
 type LLMGatewayCompletionConfig = {
@@ -198,18 +170,17 @@ type LLMGatewayCompletionConfig = {
     fetch?: typeof fetch;
     extraBody?: Record<string, unknown>;
 };
-declare class LLMGatewayCompletionLanguageModel implements LanguageModelV2 {
-    readonly specificationVersion: "v2";
+declare class LLMGatewayCompletionLanguageModel implements LanguageModelV3 {
+    readonly specificationVersion: "v3";
     readonly provider = "llmgateway";
     readonly modelId: LLMGatewayChatModelId;
     readonly supportedUrls: Record<string, RegExp[]>;
-    readonly defaultObjectGenerationMode: undefined;
     readonly settings: LLMGatewayCompletionSettings;
     private readonly config;
     constructor(modelId: LLMGatewayChatModelId, settings: LLMGatewayCompletionSettings, config: LLMGatewayCompletionConfig);
     private getArgs;
-    doGenerate(options: LanguageModelV2CallOptions): Promise<Awaited<ReturnType<LanguageModelV2['doGenerate']>>>;
-    doStream(options: LanguageModelV2CallOptions): Promise<Awaited<ReturnType<LanguageModelV2['doStream']>>>;
+    doGenerate(options: LanguageModelV3CallOptions): Promise<Awaited<ReturnType<LanguageModelV3['doGenerate']>>>;
+    doStream(options: LanguageModelV3CallOptions): Promise<Awaited<ReturnType<LanguageModelV3['doStream']>>>;
 }
 
 type LLMGatewayImageConfig = {
@@ -240,7 +211,7 @@ declare class LLMGatewayImageModel implements ImageModelV3 {
     }>;
 }
 
-interface LLMGatewayProvider extends LanguageModelV2 {
+interface LLMGatewayProvider {
     (modelId: LLMGatewayChatModelId, settings?: LLMGatewayCompletionSettings): LLMGatewayCompletionLanguageModel;
     (modelId: LLMGatewayChatModelId, settings?: LLMGatewayChatSettings): LLMGatewayChatLanguageModel;
     languageModel(modelId: LLMGatewayChatModelId, settings?: LLMGatewayCompletionSettings): LLMGatewayCompletionLanguageModel;

package/dist/index.js

@@ -2252,16 +2252,16 @@ var llmgatewayFailedResponseHandler = createJsonErrorResponseHandler({
 function mapLLMGatewayFinishReason(finishReason) {
   switch (finishReason) {
     case "stop":
-      return "stop";
+      return { unified: "stop", raw: finishReason };
     case "length":
-      return "length";
+      return { unified: "length", raw: finishReason };
     case "content_filter":
-      return "content-filter";
+      return { unified: "content-filter", raw: finishReason };
     case "function_call":
     case "tool_calls":
-      return "tool-calls";
+      return { unified: "tool-calls", raw: finishReason };
     default:
-      return "unknown";
+      return { unified: "other", raw: finishReason != null ? finishReason : void 0 };
   }
 }
 
@@ -2464,6 +2464,7 @@ function convertToLLMGatewayChatMessages(prompt) {
       }
       case "tool": {
         for (const toolResponse of content) {
+          if (toolResponse.type !== "tool-result") continue;
           const content2 = getToolResultContent(toolResponse);
           messages.push({
             role: "tool",
@@ -2482,7 +2483,14 @@ function convertToLLMGatewayChatMessages(prompt) {
   return messages;
 }
 function getToolResultContent(input) {
-  return input.output.type === "text" ? input.output.value : JSON.stringify(input.output.value);
+  switch (input.output.type) {
+    case "text":
+      return input.output.value;
+    case "json":
+      return JSON.stringify(input.output.value);
+    default:
+      return JSON.stringify(input.output);
+  }
 }
 
 // src/chat/get-tool-choice.ts
@@ -2652,9 +2660,8 @@ var LLMGatewayStreamChatCompletionChunkSchema = import_v45.z.union([
 // src/chat/index.ts
 var LLMGatewayChatLanguageModel = class {
   constructor(modelId, settings, config) {
-    this.specificationVersion = "v2";
+    this.specificationVersion = "v3";
     this.provider = "llmgateway";
-    this.defaultObjectGenerationMode = "tool";
     this.supportedUrls = {
       "image/*": [
         /^data:image\/[a-zA-Z]+;base64,/,
@@ -2743,7 +2750,7 @@ var LLMGatewayChatLanguageModel = class {
     return baseArgs;
   }
   async doGenerate(options) {
-    var _a16, _b16, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w, _x, _y, _z;
+    var _a16, _b16, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w, _x, _y;
     const providerOptions = options.providerOptions || {};
     const llmgatewayOptions = providerOptions.llmgateway || {};
     const args = __spreadValues(__spreadValues({}, this.getArgs(options)), llmgatewayOptions);
@@ -2767,19 +2774,31 @@ var LLMGatewayChatLanguageModel = class {
       throw new Error("No choice in response");
     }
     const usageInfo = response.usage ? {
-      inputTokens: (_b16 = response.usage.prompt_tokens) != null ? _b16 : 0,
-      outputTokens: (_c = response.usage.completion_tokens) != null ? _c : 0,
-      totalTokens: ((_d = response.usage.prompt_tokens) != null ? _d : 0) + ((_e = response.usage.completion_tokens) != null ? _e : 0),
-      reasoningTokens: (_g = (_f = response.usage.completion_tokens_details) == null ? void 0 : _f.reasoning_tokens) != null ? _g : 0,
-      cachedInputTokens: (_i = (_h = response.usage.prompt_tokens_details) == null ? void 0 : _h.cached_tokens) != null ? _i : 0
+      inputTokens: {
+        total: (_b16 = response.usage.prompt_tokens) != null ? _b16 : void 0,
+        noCache: void 0,
+        cacheRead: (_d = (_c = response.usage.prompt_tokens_details) == null ? void 0 : _c.cached_tokens) != null ? _d : void 0,
+        cacheWrite: void 0
+      },
+      outputTokens: {
+        total: (_e = response.usage.completion_tokens) != null ? _e : void 0,
+        text: void 0,
+        reasoning: (_g = (_f = response.usage.completion_tokens_details) == null ? void 0 : _f.reasoning_tokens) != null ? _g : void 0
+      }
     } : {
-      inputTokens: 0,
-      outputTokens: 0,
-      totalTokens: 0,
-      reasoningTokens: 0,
-      cachedInputTokens: 0
+      inputTokens: {
+        total: void 0,
+        noCache: void 0,
+        cacheRead: void 0,
+        cacheWrite: void 0
+      },
+      outputTokens: {
+        total: void 0,
+        text: void 0,
+        reasoning: void 0
+      }
     };
-    const reasoningDetails = (_j = choice.message.reasoning_details) != null ? _j : [];
+    const reasoningDetails = (_h = choice.message.reasoning_details) != null ? _h : [];
     const reasoning = reasoningDetails.length > 0 ? reasoningDetails.map((detail) => {
       switch (detail.type) {
         case "reasoning.text" /* Text */: {
@@ -2835,7 +2854,7 @@ var LLMGatewayChatLanguageModel = class {
       for (const toolCall of choice.message.tool_calls) {
         content.push({
           type: "tool-call",
-          toolCallId: (_k = toolCall.id) != null ? _k : generateId(),
+          toolCallId: (_i = toolCall.id) != null ? _i : generateId(),
           toolName: toolCall.function.name,
           input: toolCall.function.arguments
         });
@@ -2876,18 +2895,18 @@ var LLMGatewayChatLanguageModel = class {
       providerMetadata: includeUsageAccounting ? {
         llmgateway: {
           usage: {
-            promptTokens: (_l = usageInfo.inputTokens) != null ? _l : 0,
-            completionTokens: (_m = usageInfo.outputTokens) != null ? _m : 0,
-            totalTokens: (_n = usageInfo.totalTokens) != null ? _n : 0,
-            cost: typeof ((_o = response.usage) == null ? void 0 : _o.cost) === "number" ? response.usage.cost : (_q = (_p = response.usage) == null ? void 0 : _p.cost) == null ? void 0 : _q.total_cost,
+            promptTokens: (_j = usageInfo.inputTokens.total) != null ? _j : 0,
+            completionTokens: (_k = usageInfo.outputTokens.total) != null ? _k : 0,
+            totalTokens: ((_l = usageInfo.inputTokens.total) != null ? _l : 0) + ((_m = usageInfo.outputTokens.total) != null ? _m : 0),
+            cost: typeof ((_n = response.usage) == null ? void 0 : _n.cost) === "number" ? response.usage.cost : (_p = (_o = response.usage) == null ? void 0 : _o.cost) == null ? void 0 : _p.total_cost,
             promptTokensDetails: {
-              cachedTokens: (_t = (_s = (_r = response.usage) == null ? void 0 : _r.prompt_tokens_details) == null ? void 0 : _s.cached_tokens) != null ? _t : 0
+              cachedTokens: (_s = (_r = (_q = response.usage) == null ? void 0 : _q.prompt_tokens_details) == null ? void 0 : _r.cached_tokens) != null ? _s : 0
             },
             completionTokensDetails: {
-              reasoningTokens: (_w = (_v = (_u = response.usage) == null ? void 0 : _u.completion_tokens_details) == null ? void 0 : _v.reasoning_tokens) != null ? _w : 0
+              reasoningTokens: (_v = (_u = (_t = response.usage) == null ? void 0 : _t.completion_tokens_details) == null ? void 0 : _u.reasoning_tokens) != null ? _v : 0
             },
             costDetails: {
-              upstreamInferenceCost: (_z = (_y = (_x = response.usage) == null ? void 0 : _x.cost_details) == null ? void 0 : _y.upstream_inference_cost) != null ? _z : 0
+              upstreamInferenceCost: (_y = (_x = (_w = response.usage) == null ? void 0 : _w.cost_details) == null ? void 0 : _x.upstream_inference_cost) != null ? _y : 0
             }
           }
         }
@@ -2923,13 +2942,19 @@ var LLMGatewayChatLanguageModel = class {
       fetch: this.config.fetch
     });
     const toolCalls = [];
-    let finishReason = "other";
+    let finishReason = { unified: "other", raw: void 0 };
     const usage = {
-      inputTokens: Number.NaN,
-      outputTokens: Number.NaN,
-      totalTokens: Number.NaN,
-      reasoningTokens: Number.NaN,
-      cachedInputTokens: Number.NaN
+      inputTokens: {
+        total: void 0,
+        noCache: void 0,
+        cacheRead: void 0,
+        cacheWrite: void 0
+      },
+      outputTokens: {
+        total: void 0,
+        text: void 0,
+        reasoning: void 0
+      }
     };
     const llmgatewayUsage = {};
     let textStarted = false;
@@ -2943,13 +2968,13 @@ var LLMGatewayChatLanguageModel = class {
           transform(chunk, controller) {
             var _a16, _b16, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q;
             if (!chunk.success) {
-              finishReason = "error";
+              finishReason = { unified: "error", raw: void 0 };
               controller.enqueue({ type: "error", error: chunk.error });
               return;
             }
             const value = chunk.value;
             if ("error" in value) {
-              finishReason = "error";
+              finishReason = { unified: "error", raw: void 0 };
               controller.enqueue({ type: "error", error: value.error });
               return;
             }
@@ -2967,13 +2992,12 @@ var LLMGatewayChatLanguageModel = class {
               });
             }
             if (value.usage != null) {
-              usage.inputTokens = value.usage.prompt_tokens;
-              usage.outputTokens = value.usage.completion_tokens;
-              usage.totalTokens = value.usage.prompt_tokens + value.usage.completion_tokens;
+              usage.inputTokens.total = value.usage.prompt_tokens;
+              usage.outputTokens.total = value.usage.completion_tokens;
               llmgatewayUsage.promptTokens = value.usage.prompt_tokens;
               if (value.usage.prompt_tokens_details) {
                 const cachedInputTokens = (_a16 = value.usage.prompt_tokens_details.cached_tokens) != null ? _a16 : 0;
-                usage.cachedInputTokens = cachedInputTokens;
+                usage.inputTokens.cacheRead = cachedInputTokens;
                 llmgatewayUsage.promptTokensDetails = {
                   cachedTokens: cachedInputTokens
                 };
@@ -2981,7 +3005,7 @@ var LLMGatewayChatLanguageModel = class {
               llmgatewayUsage.completionTokens = value.usage.completion_tokens;
               if (value.usage.completion_tokens_details) {
                 const reasoningTokens = (_b16 = value.usage.completion_tokens_details.reasoning_tokens) != null ? _b16 : 0;
-                usage.reasoningTokens = reasoningTokens;
+                usage.outputTokens.reasoning = reasoningTokens;
                 llmgatewayUsage.completionTokensDetails = {
                   reasoningTokens
                 };
@@ -3164,7 +3188,7 @@ var LLMGatewayChatLanguageModel = class {
           },
           flush(controller) {
             var _a16;
-            if (finishReason === "tool-calls") {
+            if (finishReason.unified === "tool-calls") {
               for (const toolCall of toolCalls) {
                 if (toolCall && !toolCall.sent) {
                   controller.enqueue({
@@ -3203,7 +3227,6 @@ var LLMGatewayChatLanguageModel = class {
           }
         })
       ),
-      warnings: [],
       request: { body: args },
       response: { headers: responseHeaders }
     };
@@ -3355,7 +3378,7 @@ var LLMGatewayCompletionChunkSchema = import_v46.z.union([
 // src/completion/index.ts
 var LLMGatewayCompletionLanguageModel = class {
   constructor(modelId, settings, config) {
-    this.specificationVersion = "v2";
+    this.specificationVersion = "v3";
     this.provider = "llmgateway";
     this.supportedUrls = {
       "image/*": [
@@ -3365,7 +3388,6 @@ var LLMGatewayCompletionLanguageModel = class {
       "text/*": [/^data:text\//, /^https?:\/\/.+$/],
       "application/*": [/^data:application\//, /^https?:\/\/.+$/]
     };
-    this.defaultObjectGenerationMode = void 0;
     this.modelId = modelId;
     this.settings = settings;
     this.config = config;
@@ -3425,7 +3447,7 @@ var LLMGatewayCompletionLanguageModel = class {
     }, this.config.extraBody), this.settings.extraBody);
   }
   async doGenerate(options) {
-    var _a16, _b16, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o;
+    var _a16, _b16, _c, _d, _e, _f, _g, _h, _i, _j, _k;
     const providerOptions = options.providerOptions || {};
     const llmgatewayOptions = providerOptions.llmgateway || {};
     const args = __spreadValues(__spreadValues({}, this.getArgs(options)), llmgatewayOptions);
@@ -3459,11 +3481,17 @@ var LLMGatewayCompletionLanguageModel = class {
       ],
       finishReason: mapLLMGatewayFinishReason(choice.finish_reason),
       usage: {
-        inputTokens: (_c = (_b16 = response.usage) == null ? void 0 : _b16.prompt_tokens) != null ? _c : 0,
-        outputTokens: (_e = (_d = response.usage) == null ? void 0 : _d.completion_tokens) != null ? _e : 0,
-        totalTokens: ((_g = (_f = response.usage) == null ? void 0 : _f.prompt_tokens) != null ? _g : 0) + ((_i = (_h = response.usage) == null ? void 0 : _h.completion_tokens) != null ? _i : 0),
-        reasoningTokens: (_l = (_k = (_j = response.usage) == null ? void 0 : _j.completion_tokens_details) == null ? void 0 : _k.reasoning_tokens) != null ? _l : 0,
-        cachedInputTokens: (_o = (_n = (_m = response.usage) == null ? void 0 : _m.prompt_tokens_details) == null ? void 0 : _n.cached_tokens) != null ? _o : 0
+        inputTokens: {
+          total: (_c = (_b16 = response.usage) == null ? void 0 : _b16.prompt_tokens) != null ? _c : void 0,
+          noCache: void 0,
+          cacheRead: (_f = (_e = (_d = response.usage) == null ? void 0 : _d.prompt_tokens_details) == null ? void 0 : _e.cached_tokens) != null ? _f : void 0,
+          cacheWrite: void 0
+        },
+        outputTokens: {
+          total: (_h = (_g = response.usage) == null ? void 0 : _g.completion_tokens) != null ? _h : void 0,
+          text: void 0,
+          reasoning: (_k = (_j = (_i = response.usage) == null ? void 0 : _i.completion_tokens_details) == null ? void 0 : _j.reasoning_tokens) != null ? _k : void 0
+        }
       },
       warnings: [],
       response: {
@@ -3493,13 +3521,19 @@ var LLMGatewayCompletionLanguageModel = class {
       abortSignal: options.abortSignal,
       fetch: this.config.fetch
     });
-    let finishReason = "other";
+    let finishReason = { unified: "other", raw: void 0 };
     const usage = {
-      inputTokens: Number.NaN,
-      outputTokens: Number.NaN,
-      totalTokens: Number.NaN,
-      reasoningTokens: Number.NaN,
-      cachedInputTokens: Number.NaN
+      inputTokens: {
+        total: void 0,
+        noCache: void 0,
+        cacheRead: void 0,
+        cacheWrite: void 0
+      },
+      outputTokens: {
+        total: void 0,
+        text: void 0,
+        reasoning: void 0
+      }
     };
     const llmgatewayUsage = {};
     return {
@@ -3508,24 +3542,23 @@ var LLMGatewayCompletionLanguageModel = class {
           transform(chunk, controller) {
             var _a16, _b16, _c;
             if (!chunk.success) {
-              finishReason = "error";
+              finishReason = { unified: "error", raw: void 0 };
               controller.enqueue({ type: "error", error: chunk.error });
               return;
             }
             const value = chunk.value;
             if ("error" in value) {
-              finishReason = "error";
+              finishReason = { unified: "error", raw: void 0 };
               controller.enqueue({ type: "error", error: value.error });
               return;
             }
             if (value.usage != null) {
-              usage.inputTokens = value.usage.prompt_tokens;
-              usage.outputTokens = value.usage.completion_tokens;
-              usage.totalTokens = value.usage.prompt_tokens + value.usage.completion_tokens;
+              usage.inputTokens.total = value.usage.prompt_tokens;
+              usage.outputTokens.total = value.usage.completion_tokens;
               llmgatewayUsage.promptTokens = value.usage.prompt_tokens;
               if (value.usage.prompt_tokens_details) {
                 const cachedInputTokens = (_a16 = value.usage.prompt_tokens_details.cached_tokens) != null ? _a16 : 0;
-                usage.cachedInputTokens = cachedInputTokens;
+                usage.inputTokens.cacheRead = cachedInputTokens;
                 llmgatewayUsage.promptTokensDetails = {
                   cachedTokens: cachedInputTokens
                 };
@@ -3533,7 +3566,7 @@ var LLMGatewayCompletionLanguageModel = class {
               llmgatewayUsage.completionTokens = value.usage.completion_tokens;
               if (value.usage.completion_tokens_details) {
                 const reasoningTokens = (_b16 = value.usage.completion_tokens_details.reasoning_tokens) != null ? _b16 : 0;
-                usage.reasoningTokens = reasoningTokens;
+                usage.outputTokens.reasoning = reasoningTokens;
                 llmgatewayUsage.completionTokensDetails = {
                   reasoningTokens
                 };