@ai-sdk/xai 2.0.60 → 2.0.61

package/CHANGELOG.md

@@ -1,5 +1,11 @@
 # @ai-sdk/xai
 
+## 2.0.61
+
+### Patch Changes
+
+- c80d906: fix(provider/xai): handle inconsistent cached token reporting
+
 ## 2.0.60
 
 ### Patch Changes

package/dist/index.js

@@ -42,6 +42,30 @@ var import_provider3 = require("@ai-sdk/provider");
 var import_provider_utils3 = require("@ai-sdk/provider-utils");
 var import_v43 = require("zod/v4");
 
+// src/convert-xai-chat-usage.ts
+function convertXaiChatUsage(usage) {
+  var _a, _b, _c, _d;
+  if (usage == null) {
+    return {
+      inputTokens: void 0,
+      outputTokens: void 0,
+      totalTokens: void 0,
+      reasoningTokens: void 0,
+      cachedInputTokens: void 0
+    };
+  }
+  const cacheReadTokens = (_b = (_a = usage.prompt_tokens_details) == null ? void 0 : _a.cached_tokens) != null ? _b : 0;
+  const reasoningTokens = (_d = (_c = usage.completion_tokens_details) == null ? void 0 : _c.reasoning_tokens) != null ? _d : 0;
+  const promptTokensIncludesCached = cacheReadTokens <= usage.prompt_tokens;
+  return {
+    inputTokens: promptTokensIncludesCached ? usage.prompt_tokens : usage.prompt_tokens + cacheReadTokens,
+    outputTokens: usage.completion_tokens + reasoningTokens,
+    totalTokens: usage.total_tokens,
+    reasoningTokens: reasoningTokens || void 0,
+    cachedInputTokens: cacheReadTokens || void 0
+  };
+}
+
 // src/convert-to-xai-chat-messages.ts
 var import_provider = require("@ai-sdk/provider");
 var import_provider_utils = require("@ai-sdk/provider-utils");
@@ -468,7 +492,7 @@ var XaiChatLanguageModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n;
+    var _a;
     const { args: body, warnings } = await this.getArgs(options);
     const url = `${(_a = this.config.baseURL) != null ? _a : "https://api.x.ai/v1"}/chat/completions`;
     const {
@@ -548,13 +572,7 @@ var XaiChatLanguageModel = class {
     return {
       content,
       finishReason: mapXaiFinishReason(choice.finish_reason),
-      usage: {
-        inputTokens: (_b = response.usage) == null ? void 0 : _b.prompt_tokens,
-        outputTokens: ((_d = (_c = response.usage) == null ? void 0 : _c.completion_tokens) != null ? _d : 0) + ((_g = (_f = (_e = response.usage) == null ? void 0 : _e.completion_tokens_details) == null ? void 0 : _f.reasoning_tokens) != null ? _g : 0),
-        totalTokens: (_h = response.usage) == null ? void 0 : _h.total_tokens,
-        reasoningTokens: (_k = (_j = (_i = response.usage) == null ? void 0 : _i.completion_tokens_details) == null ? void 0 : _j.reasoning_tokens) != null ? _k : void 0,
-        cachedInputTokens: (_n = (_m = (_l = response.usage) == null ? void 0 : _l.prompt_tokens_details) == null ? void 0 : _m.cached_tokens) != null ? _n : void 0
-      },
+      usage: convertXaiChatUsage(response.usage),
       request: { body },
       response: {
         ...getResponseMetadata(response),
@@ -638,7 +656,6 @@ var XaiChatLanguageModel = class {
             controller.enqueue({ type: "stream-start", warnings });
           },
           transform(chunk, controller) {
-            var _a2, _b, _c, _d, _e, _f, _g;
             if (options.includeRawChunks) {
               controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
             }
@@ -665,11 +682,12 @@ var XaiChatLanguageModel = class {
               }
             }
             if (value.usage != null) {
-              usage.inputTokens = value.usage.prompt_tokens;
-              usage.outputTokens = ((_a2 = value.usage.completion_tokens) != null ? _a2 : 0) + ((_c = (_b = value.usage.completion_tokens_details) == null ? void 0 : _b.reasoning_tokens) != null ? _c : 0);
-              usage.totalTokens = value.usage.total_tokens;
-              usage.reasoningTokens = (_e = (_d = value.usage.completion_tokens_details) == null ? void 0 : _d.reasoning_tokens) != null ? _e : void 0;
-              usage.cachedInputTokens = (_g = (_f = value.usage.prompt_tokens_details) == null ? void 0 : _f.cached_tokens) != null ? _g : void 0;
+              const converted = convertXaiChatUsage(value.usage);
+              usage.inputTokens = converted.inputTokens;
+              usage.outputTokens = converted.outputTokens;
+              usage.totalTokens = converted.totalTokens;
+              usage.reasoningTokens = converted.reasoningTokens;
+              usage.cachedInputTokens = converted.cachedInputTokens;
             }
             const choice = value.choices[0];
             if ((choice == null ? void 0 : choice.finish_reason) != null) {
@@ -864,6 +882,28 @@ var xaiStreamErrorSchema = import_v43.z.object({
 // src/responses/xai-responses-language-model.ts
 var import_provider_utils8 = require("@ai-sdk/provider-utils");
 
+// src/responses/convert-xai-responses-usage.ts
+function convertXaiResponsesUsage(usage) {
+  var _a, _b, _c, _d;
+  if (usage == null) {
+    return {
+      inputTokens: 0,
+      outputTokens: 0,
+      totalTokens: 0
+    };
+  }
+  const cacheReadTokens = (_b = (_a = usage.input_tokens_details) == null ? void 0 : _a.cached_tokens) != null ? _b : 0;
+  const reasoningTokens = (_d = (_c = usage.output_tokens_details) == null ? void 0 : _c.reasoning_tokens) != null ? _d : 0;
+  const inputTokensIncludesCached = cacheReadTokens <= usage.input_tokens;
+  return {
+    inputTokens: inputTokensIncludesCached ? usage.input_tokens : usage.input_tokens + cacheReadTokens,
+    outputTokens: usage.output_tokens,
+    totalTokens: usage.total_tokens,
+    reasoningTokens: reasoningTokens || void 0,
+    cachedInputTokens: cacheReadTokens || void 0
+  };
+}
+
 // src/responses/xai-responses-api.ts
 var import_v44 = require("zod/v4");
 var annotationSchema = import_v44.z.union([
@@ -1717,7 +1757,7 @@ var XaiResponsesLanguageModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e, _f, _g, _h, _i;
+    var _a, _b, _c, _d, _e, _f, _g;
     const {
       args: body,
       warnings,
@@ -1840,17 +1880,7 @@ var XaiResponsesLanguageModel = class {
     return {
       content,
       finishReason: mapXaiResponsesFinishReason(response.status),
-      usage: response.usage ? {
-        inputTokens: response.usage.input_tokens,
-        outputTokens: response.usage.output_tokens,
-        totalTokens: response.usage.total_tokens,
-        reasoningTokens: (_h = response.usage.output_tokens_details) == null ? void 0 : _h.reasoning_tokens,
-        cachedInputTokens: (_i = response.usage.input_tokens_details) == null ? void 0 : _i.cached_tokens
-      } : {
-        inputTokens: 0,
-        outputTokens: 0,
-        totalTokens: 0
-      },
+      usage: convertXaiResponsesUsage(response.usage),
       request: { body },
       response: {
         ...getResponseMetadata(response),
@@ -1903,7 +1933,7 @@ var XaiResponsesLanguageModel = class {
             controller.enqueue({ type: "stream-start", warnings });
           },
           transform(chunk, controller) {
-            var _a2, _b, _c, _d, _e, _f, _g, _h, _i, _j;
+            var _a2, _b, _c, _d, _e, _f, _g, _h;
             if (options.includeRawChunks) {
               controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
             }
@@ -2028,11 +2058,12 @@ var XaiResponsesLanguageModel = class {
             if (event.type === "response.done" || event.type === "response.completed") {
               const response2 = event.response;
               if (response2.usage) {
-                usage.inputTokens = response2.usage.input_tokens;
-                usage.cachedInputTokens = (_c = response2.usage.input_tokens_details) == null ? void 0 : _c.cached_tokens;
-                usage.outputTokens = response2.usage.output_tokens;
-                usage.totalTokens = response2.usage.total_tokens;
-                usage.reasoningTokens = (_d = response2.usage.output_tokens_details) == null ? void 0 : _d.reasoning_tokens;
+                const converted = convertXaiResponsesUsage(response2.usage);
+                usage.inputTokens = converted.inputTokens;
+                usage.outputTokens = converted.outputTokens;
+                usage.totalTokens = converted.totalTokens;
+                usage.reasoningTokens = converted.reasoningTokens;
+                usage.cachedInputTokens = converted.cachedInputTokens;
               }
               if (response2.status) {
                 finishReason = mapXaiResponsesFinishReason(response2.status);
@@ -2096,15 +2127,15 @@ var XaiResponsesLanguageModel = class {
                   "x_semantic_search",
                   "x_thread_fetch"
                 ];
-                let toolName = (_e = part.name) != null ? _e : "";
-                if (webSearchSubTools.includes((_f = part.name) != null ? _f : "") || part.type === "web_search_call") {
+                let toolName = (_c = part.name) != null ? _c : "";
+                if (webSearchSubTools.includes((_d = part.name) != null ? _d : "") || part.type === "web_search_call") {
                   toolName = webSearchToolName != null ? webSearchToolName : "web_search";
-                } else if (xSearchSubTools.includes((_g = part.name) != null ? _g : "") || part.type === "x_search_call") {
+                } else if (xSearchSubTools.includes((_e = part.name) != null ? _e : "") || part.type === "x_search_call") {
                   toolName = xSearchToolName != null ? xSearchToolName : "x_search";
                 } else if (part.name === "code_execution" || part.type === "code_interpreter_call" || part.type === "code_execution_call") {
                   toolName = codeExecutionToolName != null ? codeExecutionToolName : "code_execution";
                 }
-                const toolInput = part.type === "custom_tool_call" ? (_h = part.input) != null ? _h : "" : (_i = part.arguments) != null ? _i : "";
+                const toolInput = part.type === "custom_tool_call" ? (_f = part.input) != null ? _f : "" : (_g = part.arguments) != null ? _g : "";
                 const shouldEmit = part.type === "custom_tool_call" ? event.type === "response.output_item.done" : !seenToolCalls.has(part.id);
                 if (shouldEmit && !seenToolCalls.has(part.id)) {
                   seenToolCalls.add(part.id);
@@ -2157,7 +2188,7 @@ var XaiResponsesLanguageModel = class {
                           sourceType: "url",
                           id: self.config.generateId(),
                           url: annotation.url,
-                          title: (_j = annotation.title) != null ? _j : annotation.url
+                          title: (_h = annotation.title) != null ? _h : annotation.url
                         });
                       }
                     }
@@ -2265,7 +2296,7 @@ var xaiTools = {
 };
 
 // src/version.ts
-var VERSION = true ? "2.0.60" : "0.0.0-test";
+var VERSION = true ? "2.0.61" : "0.0.0-test";
 
 // src/xai-provider.ts
 var xaiErrorStructure = {