@ai-sdk/openai-compatible 2.0.0-beta.44 → 2.0.0-beta.46

package/CHANGELOG.md

@@ -1,5 +1,22 @@
 # @ai-sdk/openai-compatible
 
+## 2.0.0-beta.46
+
+### Patch Changes
+
+- 3bd2689: feat: extended token usage
+- Updated dependencies [3bd2689]
+  - @ai-sdk/provider@3.0.0-beta.26
+  - @ai-sdk/provider-utils@4.0.0-beta.45
+
+## 2.0.0-beta.45
+
+### Patch Changes
+
+- Updated dependencies [53f3368]
+  - @ai-sdk/provider@3.0.0-beta.25
+  - @ai-sdk/provider-utils@4.0.0-beta.44
+
 ## 2.0.0-beta.44
 
 ### Patch Changes

package/dist/index.d.mts

@@ -1,14 +1,6 @@
 import { SharedV3ProviderMetadata, LanguageModelV3, EmbeddingModelV3, ImageModelV3, ProviderV3 } from '@ai-sdk/provider';
 import { FetchFunction } from '@ai-sdk/provider-utils';
-import { z, ZodType } from 'zod/v4';
-
-type OpenAICompatibleChatModelId = string;
-declare const openaiCompatibleProviderOptions: z.ZodObject<{
-    user: z.ZodOptional<z.ZodString>;
-    reasoningEffort: z.ZodOptional<z.ZodString>;
-    textVerbosity: z.ZodOptional<z.ZodString>;
-}, z.core.$strip>;
-type OpenAICompatibleProviderOptions = z.infer<typeof openaiCompatibleProviderOptions>;
+import { ZodType, z } from 'zod/v4';
 
 declare const openaiCompatibleErrorDataSchema: z.ZodObject<{
     error: z.ZodObject<{
@@ -25,6 +17,14 @@ type ProviderErrorStructure<T> = {
     isRetryable?: (response: Response, error?: T) => boolean;
 };
 
+type OpenAICompatibleChatModelId = string;
+declare const openaiCompatibleProviderOptions: z.ZodObject<{
+    user: z.ZodOptional<z.ZodString>;
+    reasoningEffort: z.ZodOptional<z.ZodString>;
+    textVerbosity: z.ZodOptional<z.ZodString>;
+}, z.core.$strip>;
+type OpenAICompatibleProviderOptions = z.infer<typeof openaiCompatibleProviderOptions>;
+
 /**
 Extracts provider-specific metadata from API responses.
 Used to standardize metadata handling across different LLM providers while allowing

package/dist/index.d.ts

@@ -1,14 +1,6 @@
 import { SharedV3ProviderMetadata, LanguageModelV3, EmbeddingModelV3, ImageModelV3, ProviderV3 } from '@ai-sdk/provider';
 import { FetchFunction } from '@ai-sdk/provider-utils';
-import { z, ZodType } from 'zod/v4';
-
-type OpenAICompatibleChatModelId = string;
-declare const openaiCompatibleProviderOptions: z.ZodObject<{
-    user: z.ZodOptional<z.ZodString>;
-    reasoningEffort: z.ZodOptional<z.ZodString>;
-    textVerbosity: z.ZodOptional<z.ZodString>;
-}, z.core.$strip>;
-type OpenAICompatibleProviderOptions = z.infer<typeof openaiCompatibleProviderOptions>;
+import { ZodType, z } from 'zod/v4';
 
 declare const openaiCompatibleErrorDataSchema: z.ZodObject<{
     error: z.ZodObject<{
@@ -25,6 +17,14 @@ type ProviderErrorStructure<T> = {
     isRetryable?: (response: Response, error?: T) => boolean;
 };
 
+type OpenAICompatibleChatModelId = string;
+declare const openaiCompatibleProviderOptions: z.ZodObject<{
+    user: z.ZodOptional<z.ZodString>;
+    reasoningEffort: z.ZodOptional<z.ZodString>;
+    textVerbosity: z.ZodOptional<z.ZodString>;
+}, z.core.$strip>;
+type OpenAICompatibleProviderOptions = z.infer<typeof openaiCompatibleProviderOptions>;
+
 /**
 Extracts provider-specific metadata from API responses.
 Used to standardize metadata handling across different LLM providers while allowing

package/dist/index.js

@@ -34,6 +34,63 @@ var import_provider3 = require("@ai-sdk/provider");
 var import_provider_utils2 = require("@ai-sdk/provider-utils");
 var import_v43 = require("zod/v4");
 
+// src/openai-compatible-error.ts
+var import_v4 = require("zod/v4");
+var openaiCompatibleErrorDataSchema = import_v4.z.object({
+  error: import_v4.z.object({
+    message: import_v4.z.string(),
+    // The additional information below is handled loosely to support
+    // OpenAI-compatible providers that have slightly different error
+    // responses:
+    type: import_v4.z.string().nullish(),
+    param: import_v4.z.any().nullish(),
+    code: import_v4.z.union([import_v4.z.string(), import_v4.z.number()]).nullish()
+  })
+});
+var defaultOpenAICompatibleErrorStructure = {
+  errorSchema: openaiCompatibleErrorDataSchema,
+  errorToMessage: (data) => data.error.message
+};
+
+// src/chat/convert-openai-compatible-chat-usage.ts
+function convertOpenAICompatibleChatUsage(usage) {
+  var _a, _b, _c, _d, _e, _f;
+  if (usage == null) {
+    return {
+      inputTokens: {
+        total: void 0,
+        noCache: void 0,
+        cacheRead: void 0,
+        cacheWrite: void 0
+      },
+      outputTokens: {
+        total: void 0,
+        text: void 0,
+        reasoning: void 0
+      },
+      raw: void 0
+    };
+  }
+  const promptTokens = (_a = usage.prompt_tokens) != null ? _a : 0;
+  const completionTokens = (_b = usage.completion_tokens) != null ? _b : 0;
+  const cacheReadTokens = (_d = (_c = usage.prompt_tokens_details) == null ? void 0 : _c.cached_tokens) != null ? _d : 0;
+  const reasoningTokens = (_f = (_e = usage.completion_tokens_details) == null ? void 0 : _e.reasoning_tokens) != null ? _f : 0;
+  return {
+    inputTokens: {
+      total: promptTokens,
+      noCache: promptTokens - cacheReadTokens,
+      cacheRead: cacheReadTokens,
+      cacheWrite: void 0
+    },
+    outputTokens: {
+      total: completionTokens,
+      text: completionTokens - reasoningTokens,
+      reasoning: reasoningTokens
+    },
+    raw: usage
+  };
+}
+
 // src/chat/convert-to-openai-compatible-chat-messages.ts
 var import_provider = require("@ai-sdk/provider");
 var import_provider_utils = require("@ai-sdk/provider-utils");
@@ -190,41 +247,23 @@ function mapOpenAICompatibleFinishReason(finishReason) {
 }
 
 // src/chat/openai-compatible-chat-options.ts
-var import_v4 = require("zod/v4");
-var openaiCompatibleProviderOptions = import_v4.z.object({
+var import_v42 = require("zod/v4");
+var openaiCompatibleProviderOptions = import_v42.z.object({
   /**
    * A unique identifier representing your end-user, which can help the provider to
    * monitor and detect abuse.
    */
-  user: import_v4.z.string().optional(),
+  user: import_v42.z.string().optional(),
   /**
    * Reasoning effort for reasoning models. Defaults to `medium`.
    */
-  reasoningEffort: import_v4.z.string().optional(),
+  reasoningEffort: import_v42.z.string().optional(),
   /**
    * Controls the verbosity of the generated text. Defaults to `medium`.
    */
-  textVerbosity: import_v4.z.string().optional()
+  textVerbosity: import_v42.z.string().optional()
 });
 
-// src/openai-compatible-error.ts
-var import_v42 = require("zod/v4");
-var openaiCompatibleErrorDataSchema = import_v42.z.object({
-  error: import_v42.z.object({
-    message: import_v42.z.string(),
-    // The additional information below is handled loosely to support
-    // OpenAI-compatible providers that have slightly different error
-    // responses:
-    type: import_v42.z.string().nullish(),
-    param: import_v42.z.any().nullish(),
-    code: import_v42.z.union([import_v42.z.string(), import_v42.z.number()]).nullish()
-  })
-});
-var defaultOpenAICompatibleErrorStructure = {
-  errorSchema: openaiCompatibleErrorDataSchema,
-  errorToMessage: (data) => data.error.message
-};
-
 // src/chat/openai-compatible-prepare-tools.ts
 var import_provider2 = require("@ai-sdk/provider");
 function prepareTools({
@@ -395,7 +434,7 @@ var OpenAICompatibleChatLanguageModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q;
+    var _a, _b, _c, _d, _e;
     const { args, warnings } = await this.getArgs({ ...options });
     const body = JSON.stringify(args);
     const {
@@ -455,13 +494,7 @@ var OpenAICompatibleChatLanguageModel = class {
     return {
       content,
       finishReason: mapOpenAICompatibleFinishReason(choice.finish_reason),
-      usage: {
-        inputTokens: (_g = (_f = responseBody.usage) == null ? void 0 : _f.prompt_tokens) != null ? _g : void 0,
-        outputTokens: (_i = (_h = responseBody.usage) == null ? void 0 : _h.completion_tokens) != null ? _i : void 0,
-        totalTokens: (_k = (_j = responseBody.usage) == null ? void 0 : _j.total_tokens) != null ? _k : void 0,
-        reasoningTokens: (_n = (_m = (_l = responseBody.usage) == null ? void 0 : _l.completion_tokens_details) == null ? void 0 : _m.reasoning_tokens) != null ? _n : void 0,
-        cachedInputTokens: (_q = (_p = (_o = responseBody.usage) == null ? void 0 : _o.prompt_tokens_details) == null ? void 0 : _p.cached_tokens) != null ? _q : void 0
-      },
+      usage: convertOpenAICompatibleChatUsage(responseBody.usage),
       providerMetadata,
       request: { body },
       response: {
@@ -498,19 +531,7 @@ var OpenAICompatibleChatLanguageModel = class {
     });
     const toolCalls = [];
     let finishReason = "unknown";
-    const usage = {
-      completionTokens: void 0,
-      completionTokensDetails: {
-        reasoningTokens: void 0,
-        acceptedPredictionTokens: void 0,
-        rejectedPredictionTokens: void 0
-      },
-      promptTokens: void 0,
-      promptTokensDetails: {
-        cachedTokens: void 0
-      },
-      totalTokens: void 0
-    };
+    let usage = void 0;
     let isFirstChunk = true;
     const providerOptionsName = this.providerOptionsName;
     let isActiveReasoning = false;
@@ -521,7 +542,6 @@ var OpenAICompatibleChatLanguageModel = class {
           start(controller) {
             controller.enqueue({ type: "stream-start", warnings });
           },
-          // TODO we lost type safety on Chunk, most likely due to the error schema. MUST FIX
           transform(chunk, controller) {
             var _a2, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m;
             if (options.includeRawChunks) {
@@ -532,13 +552,16 @@ var OpenAICompatibleChatLanguageModel = class {
               controller.enqueue({ type: "error", error: chunk.error });
               return;
             }
-            const value = chunk.value;
             metadataExtractor == null ? void 0 : metadataExtractor.processChunk(chunk.rawValue);
-            if ("error" in value) {
+            if ("error" in chunk.value) {
               finishReason = "error";
-              controller.enqueue({ type: "error", error: value.error.message });
+              controller.enqueue({
+                type: "error",
+                error: chunk.value.error.message
+              });
               return;
             }
+            const value = chunk.value;
             if (isFirstChunk) {
               isFirstChunk = false;
               controller.enqueue({
@@ -547,28 +570,7 @@ var OpenAICompatibleChatLanguageModel = class {
               });
             }
             if (value.usage != null) {
-              const {
-                prompt_tokens,
-                completion_tokens,
-                total_tokens,
-                prompt_tokens_details,
-                completion_tokens_details
-              } = value.usage;
-              usage.promptTokens = prompt_tokens != null ? prompt_tokens : void 0;
-              usage.completionTokens = completion_tokens != null ? completion_tokens : void 0;
-              usage.totalTokens = total_tokens != null ? total_tokens : void 0;
-              if ((completion_tokens_details == null ? void 0 : completion_tokens_details.reasoning_tokens) != null) {
-                usage.completionTokensDetails.reasoningTokens = completion_tokens_details == null ? void 0 : completion_tokens_details.reasoning_tokens;
-              }
-              if ((completion_tokens_details == null ? void 0 : completion_tokens_details.accepted_prediction_tokens) != null) {
-                usage.completionTokensDetails.acceptedPredictionTokens = completion_tokens_details == null ? void 0 : completion_tokens_details.accepted_prediction_tokens;
-              }
-              if ((completion_tokens_details == null ? void 0 : completion_tokens_details.rejected_prediction_tokens) != null) {
-                usage.completionTokensDetails.rejectedPredictionTokens = completion_tokens_details == null ? void 0 : completion_tokens_details.rejected_prediction_tokens;
-              }
-              if ((prompt_tokens_details == null ? void 0 : prompt_tokens_details.cached_tokens) != null) {
-                usage.promptTokensDetails.cachedTokens = prompt_tokens_details == null ? void 0 : prompt_tokens_details.cached_tokens;
-              }
+              usage = value.usage;
             }
             const choice = value.choices[0];
             if ((choice == null ? void 0 : choice.finish_reason) != null) {
@@ -690,7 +692,7 @@ var OpenAICompatibleChatLanguageModel = class {
             }
           },
           flush(controller) {
-            var _a2, _b, _c, _d, _e, _f;
+            var _a2, _b, _c, _d, _e;
             if (isActiveReasoning) {
               controller.enqueue({ type: "reasoning-end", id: "reasoning-0" });
             }
@@ -715,22 +717,16 @@ var OpenAICompatibleChatLanguageModel = class {
               [providerOptionsName]: {},
               ...metadataExtractor == null ? void 0 : metadataExtractor.buildMetadata()
             };
-            if (usage.completionTokensDetails.acceptedPredictionTokens != null) {
-              providerMetadata[providerOptionsName].acceptedPredictionTokens = usage.completionTokensDetails.acceptedPredictionTokens;
+            if (((_b = usage == null ? void 0 : usage.completion_tokens_details) == null ? void 0 : _b.accepted_prediction_tokens) != null) {
+              providerMetadata[providerOptionsName].acceptedPredictionTokens = (_c = usage == null ? void 0 : usage.completion_tokens_details) == null ? void 0 : _c.accepted_prediction_tokens;
             }
-            if (usage.completionTokensDetails.rejectedPredictionTokens != null) {
-              providerMetadata[providerOptionsName].rejectedPredictionTokens = usage.completionTokensDetails.rejectedPredictionTokens;
+            if (((_d = usage == null ? void 0 : usage.completion_tokens_details) == null ? void 0 : _d.rejected_prediction_tokens) != null) {
+              providerMetadata[providerOptionsName].rejectedPredictionTokens = (_e = usage == null ? void 0 : usage.completion_tokens_details) == null ? void 0 : _e.rejected_prediction_tokens;
             }
             controller.enqueue({
               type: "finish",
               finishReason,
-              usage: {
-                inputTokens: (_b = usage.promptTokens) != null ? _b : void 0,
-                outputTokens: (_c = usage.completionTokens) != null ? _c : void 0,
-                totalTokens: (_d = usage.totalTokens) != null ? _d : void 0,
-                reasoningTokens: (_e = usage.completionTokensDetails.reasoningTokens) != null ? _e : void 0,
-                cachedInputTokens: (_f = usage.promptTokensDetails.cachedTokens) != null ? _f : void 0
-              },
+              usage: convertOpenAICompatibleChatUsage(usage),
               providerMetadata
             });
           }
@@ -780,43 +776,78 @@ var OpenAICompatibleChatResponseSchema = import_v43.z.object({
   ),
   usage: openaiCompatibleTokenUsageSchema
 });
-var createOpenAICompatibleChatChunkSchema = (errorSchema) => import_v43.z.union([
-  import_v43.z.object({
-    id: import_v43.z.string().nullish(),
-    created: import_v43.z.number().nullish(),
-    model: import_v43.z.string().nullish(),
-    choices: import_v43.z.array(
-      import_v43.z.object({
-        delta: import_v43.z.object({
-          role: import_v43.z.enum(["assistant"]).nullish(),
-          content: import_v43.z.string().nullish(),
-          // Most openai-compatible models set `reasoning_content`, but some
-          // providers serving `gpt-oss` set `reasoning`. See #7866
-          reasoning_content: import_v43.z.string().nullish(),
-          reasoning: import_v43.z.string().nullish(),
-          tool_calls: import_v43.z.array(
-            import_v43.z.object({
-              index: import_v43.z.number(),
-              id: import_v43.z.string().nullish(),
-              function: import_v43.z.object({
-                name: import_v43.z.string().nullish(),
-                arguments: import_v43.z.string().nullish()
-              })
+var chunkBaseSchema = import_v43.z.object({
+  id: import_v43.z.string().nullish(),
+  created: import_v43.z.number().nullish(),
+  model: import_v43.z.string().nullish(),
+  choices: import_v43.z.array(
+    import_v43.z.object({
+      delta: import_v43.z.object({
+        role: import_v43.z.enum(["assistant"]).nullish(),
+        content: import_v43.z.string().nullish(),
+        // Most openai-compatible models set `reasoning_content`, but some
+        // providers serving `gpt-oss` set `reasoning`. See #7866
+        reasoning_content: import_v43.z.string().nullish(),
+        reasoning: import_v43.z.string().nullish(),
+        tool_calls: import_v43.z.array(
+          import_v43.z.object({
+            index: import_v43.z.number(),
+            id: import_v43.z.string().nullish(),
+            function: import_v43.z.object({
+              name: import_v43.z.string().nullish(),
+              arguments: import_v43.z.string().nullish()
             })
-          ).nullish()
-        }).nullish(),
-        finish_reason: import_v43.z.string().nullish()
-      })
-    ),
-    usage: openaiCompatibleTokenUsageSchema
-  }),
-  errorSchema
-]);
+          })
+        ).nullish()
+      }).nullish(),
+      finish_reason: import_v43.z.string().nullish()
+    })
+  ),
+  usage: openaiCompatibleTokenUsageSchema
+});
+var createOpenAICompatibleChatChunkSchema = (errorSchema) => import_v43.z.union([chunkBaseSchema, errorSchema]);
 
 // src/completion/openai-compatible-completion-language-model.ts
 var import_provider_utils3 = require("@ai-sdk/provider-utils");
 var import_v45 = require("zod/v4");
 
+// src/completion/convert-openai-compatible-completion-usage.ts
+function convertOpenAICompatibleCompletionUsage(usage) {
+  var _a, _b;
+  if (usage == null) {
+    return {
+      inputTokens: {
+        total: void 0,
+        noCache: void 0,
+        cacheRead: void 0,
+        cacheWrite: void 0
+      },
+      outputTokens: {
+        total: void 0,
+        text: void 0,
+        reasoning: void 0
+      },
+      raw: void 0
+    };
+  }
+  const promptTokens = (_a = usage.prompt_tokens) != null ? _a : 0;
+  const completionTokens = (_b = usage.completion_tokens) != null ? _b : 0;
+  return {
+    inputTokens: {
+      total: promptTokens,
+      noCache: promptTokens,
+      cacheRead: void 0,
+      cacheWrite: void 0
+    },
+    outputTokens: {
+      total: completionTokens,
+      text: completionTokens,
+      reasoning: void 0
+    },
+    raw: usage
+  };
+}
+
 // src/completion/convert-to-openai-compatible-completion-prompt.ts
 var import_provider4 = require("@ai-sdk/provider");
 function convertToOpenAICompatibleCompletionPrompt({
@@ -1037,7 +1068,6 @@ var OpenAICompatibleCompletionLanguageModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e, _f;
     const { args, warnings } = await this.getArgs(options);
     const {
       responseHeaders,
@@ -1064,11 +1094,7 @@ var OpenAICompatibleCompletionLanguageModel = class {
     }
     return {
       content,
-      usage: {
-        inputTokens: (_b = (_a = response.usage) == null ? void 0 : _a.prompt_tokens) != null ? _b : void 0,
-        outputTokens: (_d = (_c = response.usage) == null ? void 0 : _c.completion_tokens) != null ? _d : void 0,
-        totalTokens: (_f = (_e = response.usage) == null ? void 0 : _e.total_tokens) != null ? _f : void 0
-      },
+      usage: convertOpenAICompatibleCompletionUsage(response.usage),
       finishReason: mapOpenAICompatibleFinishReason2(choice.finish_reason),
       request: { body: args },
       response: {
@@ -1102,11 +1128,7 @@ var OpenAICompatibleCompletionLanguageModel = class {
       fetch: this.config.fetch
     });
     let finishReason = "unknown";
-    const usage = {
-      inputTokens: void 0,
-      outputTokens: void 0,
-      totalTokens: void 0
-    };
+    let usage = void 0;
     let isFirstChunk = true;
     return {
       stream: response.pipeThrough(
@@ -1115,7 +1137,6 @@ var OpenAICompatibleCompletionLanguageModel = class {
             controller.enqueue({ type: "stream-start", warnings });
           },
           transform(chunk, controller) {
-            var _a, _b, _c;
             if (options.includeRawChunks) {
               controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
             }
@@ -1142,9 +1163,7 @@ var OpenAICompatibleCompletionLanguageModel = class {
               });
             }
             if (value.usage != null) {
-              usage.inputTokens = (_a = value.usage.prompt_tokens) != null ? _a : void 0;
-              usage.outputTokens = (_b = value.usage.completion_tokens) != null ? _b : void 0;
-              usage.totalTokens = (_c = value.usage.total_tokens) != null ? _c : void 0;
+              usage = value.usage;
             }
             const choice = value.choices[0];
             if ((choice == null ? void 0 : choice.finish_reason) != null) {
@@ -1167,7 +1186,7 @@ var OpenAICompatibleCompletionLanguageModel = class {
             controller.enqueue({
               type: "finish",
               finishReason,
-              usage
+              usage: convertOpenAICompatibleCompletionUsage(usage)
             });
           }
         })
@@ -1306,6 +1325,7 @@ var OpenAICompatibleEmbeddingModel = class {
       fetch: this.config.fetch
     });
     return {
+      warnings: [],
       embeddings: response.data.map((item) => item.embedding),
       usage: response.usage ? { tokens: response.usage.prompt_tokens } : void 0,
       providerMetadata: response.providerMetadata,
@@ -1397,7 +1417,7 @@ var openaiCompatibleImageResponseSchema = import_v48.z.object({
 var import_provider_utils6 = require("@ai-sdk/provider-utils");
 
 // src/version.ts
-var VERSION = true ? "2.0.0-beta.44" : "0.0.0-test";
+var VERSION = true ? "2.0.0-beta.46" : "0.0.0-test";
 
 // src/openai-compatible-provider.ts
 function createOpenAICompatible(options) {