@ai-sdk/openai 2.0.0-canary.15 → 2.0.0-canary.17

package/dist/internal/index.mjs

@@ -529,7 +529,7 @@ var OpenAIChatLanguageModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e, _f, _g, _h, _i;
+    var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m;
     const { args: body, warnings } = await this.getArgs(options);
     const {
       responseHeaders,
@@ -567,18 +567,12 @@ var OpenAIChatLanguageModel = class {
     const completionTokenDetails = (_c = response.usage) == null ? void 0 : _c.completion_tokens_details;
     const promptTokenDetails = (_d = response.usage) == null ? void 0 : _d.prompt_tokens_details;
     const providerMetadata = { openai: {} };
-    if ((completionTokenDetails == null ? void 0 : completionTokenDetails.reasoning_tokens) != null) {
-      providerMetadata.openai.reasoningTokens = completionTokenDetails == null ? void 0 : completionTokenDetails.reasoning_tokens;
-    }
     if ((completionTokenDetails == null ? void 0 : completionTokenDetails.accepted_prediction_tokens) != null) {
       providerMetadata.openai.acceptedPredictionTokens = completionTokenDetails == null ? void 0 : completionTokenDetails.accepted_prediction_tokens;
     }
     if ((completionTokenDetails == null ? void 0 : completionTokenDetails.rejected_prediction_tokens) != null) {
       providerMetadata.openai.rejectedPredictionTokens = completionTokenDetails == null ? void 0 : completionTokenDetails.rejected_prediction_tokens;
     }
-    if ((promptTokenDetails == null ? void 0 : promptTokenDetails.cached_tokens) != null) {
-      providerMetadata.openai.cachedPromptTokens = promptTokenDetails == null ? void 0 : promptTokenDetails.cached_tokens;
-    }
     if (((_e = choice.logprobs) == null ? void 0 : _e.content) != null) {
       providerMetadata.openai.logprobs = choice.logprobs.content;
     }
@@ -587,7 +581,10 @@ var OpenAIChatLanguageModel = class {
       finishReason: mapOpenAIFinishReason(choice.finish_reason),
       usage: {
         inputTokens: (_g = (_f = response.usage) == null ? void 0 : _f.prompt_tokens) != null ? _g : void 0,
-        outputTokens: (_i = (_h = response.usage) == null ? void 0 : _h.completion_tokens) != null ? _i : void 0
+        outputTokens: (_i = (_h = response.usage) == null ? void 0 : _h.completion_tokens) != null ? _i : void 0,
+        totalTokens: (_k = (_j = response.usage) == null ? void 0 : _j.total_tokens) != null ? _k : void 0,
+        reasoningTokens: (_l = completionTokenDetails == null ? void 0 : completionTokenDetails.reasoning_tokens) != null ? _l : void 0,
+        cachedInputTokens: (_m = promptTokenDetails == null ? void 0 : promptTokenDetails.cached_tokens) != null ? _m : void 0
       },
       request: { body },
       response: {
@@ -604,8 +601,9 @@ var OpenAIChatLanguageModel = class {
     const body = {
       ...args,
       stream: true,
-      // only include stream_options when in strict compatibility mode:
-      stream_options: this.config.compatibility === "strict" ? { include_usage: true } : void 0
+      stream_options: {
+        include_usage: true
+      }
     };
     const { responseHeaders, value: response } = await postJsonToApi({
       url: this.config.url({
@@ -621,12 +619,12 @@ var OpenAIChatLanguageModel = class {
       abortSignal: options.abortSignal,
       fetch: this.config.fetch
     });
-    const { messages: rawPrompt, ...rawSettings } = args;
     const toolCalls = [];
     let finishReason = "unknown";
     const usage = {
       inputTokens: void 0,
-      outputTokens: void 0
+      outputTokens: void 0,
+      totalTokens: void 0
     };
     let isFirstChunk = true;
     const providerMetadata = { openai: {} };
@@ -637,7 +635,7 @@ var OpenAIChatLanguageModel = class {
             controller.enqueue({ type: "stream-start", warnings });
           },
           transform(chunk, controller) {
-            var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l;
+            var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w, _x;
             if (!chunk.success) {
               finishReason = "error";
               controller.enqueue({ type: "error", error: chunk.error });
@@ -657,31 +655,25 @@ var OpenAIChatLanguageModel = class {
               });
             }
             if (value.usage != null) {
-              const {
-                prompt_tokens,
-                completion_tokens,
-                prompt_tokens_details,
-                completion_tokens_details
-              } = value.usage;
-              usage.inputTokens = prompt_tokens != null ? prompt_tokens : void 0;
-              usage.outputTokens = completion_tokens != null ? completion_tokens : void 0;
-              if ((completion_tokens_details == null ? void 0 : completion_tokens_details.reasoning_tokens) != null) {
-                providerMetadata.openai.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) {
-                providerMetadata.openai.acceptedPredictionTokens = completion_tokens_details == null ? void 0 : completion_tokens_details.accepted_prediction_tokens;
+              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.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;
+              if (((_h = value.usage.completion_tokens_details) == null ? void 0 : _h.accepted_prediction_tokens) != null) {
+                providerMetadata.openai.acceptedPredictionTokens = (_i = value.usage.completion_tokens_details) == null ? void 0 : _i.accepted_prediction_tokens;
               }
-              if ((completion_tokens_details == null ? void 0 : completion_tokens_details.rejected_prediction_tokens) != null) {
-                providerMetadata.openai.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) {
-                providerMetadata.openai.cachedPromptTokens = prompt_tokens_details == null ? void 0 : prompt_tokens_details.cached_tokens;
+              if (((_j = value.usage.completion_tokens_details) == null ? void 0 : _j.rejected_prediction_tokens) != null) {
+                providerMetadata.openai.rejectedPredictionTokens = (_k = value.usage.completion_tokens_details) == null ? void 0 : _k.rejected_prediction_tokens;
               }
             }
             const choice = value.choices[0];
             if ((choice == null ? void 0 : choice.finish_reason) != null) {
               finishReason = mapOpenAIFinishReason(choice.finish_reason);
             }
+            if (((_l = choice == null ? void 0 : choice.logprobs) == null ? void 0 : _l.content) != null) {
+              providerMetadata.openai.logprobs = choice.logprobs.content;
+            }
             if ((choice == null ? void 0 : choice.delta) == null) {
               return;
             }
@@ -708,7 +700,7 @@ var OpenAIChatLanguageModel = class {
                       message: `Expected 'id' to be a string.`
                     });
                   }
-                  if (((_a = toolCallDelta.function) == null ? void 0 : _a.name) == null) {
+                  if (((_m = toolCallDelta.function) == null ? void 0 : _m.name) == null) {
                     throw new InvalidResponseDataError({
                       data: toolCallDelta,
                       message: `Expected 'function.name' to be a string.`
@@ -719,12 +711,12 @@ var OpenAIChatLanguageModel = class {
                     type: "function",
                     function: {
                       name: toolCallDelta.function.name,
-                      arguments: (_b = toolCallDelta.function.arguments) != null ? _b : ""
+                      arguments: (_n = toolCallDelta.function.arguments) != null ? _n : ""
                     },
                     hasFinished: false
                   };
                   const toolCall2 = toolCalls[index];
-                  if (((_c = toolCall2.function) == null ? void 0 : _c.name) != null && ((_d = toolCall2.function) == null ? void 0 : _d.arguments) != null) {
+                  if (((_o = toolCall2.function) == null ? void 0 : _o.name) != null && ((_p = toolCall2.function) == null ? void 0 : _p.arguments) != null) {
                     if (toolCall2.function.arguments.length > 0) {
                       controller.enqueue({
                         type: "tool-call-delta",
@@ -738,7 +730,7 @@ var OpenAIChatLanguageModel = class {
                       controller.enqueue({
                         type: "tool-call",
                         toolCallType: "function",
-                        toolCallId: (_e = toolCall2.id) != null ? _e : generateId(),
+                        toolCallId: (_q = toolCall2.id) != null ? _q : generateId(),
                         toolName: toolCall2.function.name,
                         args: toolCall2.function.arguments
                       });
@@ -751,21 +743,21 @@ var OpenAIChatLanguageModel = class {
                 if (toolCall.hasFinished) {
                   continue;
                 }
-                if (((_f = toolCallDelta.function) == null ? void 0 : _f.arguments) != null) {
-                  toolCall.function.arguments += (_h = (_g = toolCallDelta.function) == null ? void 0 : _g.arguments) != null ? _h : "";
+                if (((_r = toolCallDelta.function) == null ? void 0 : _r.arguments) != null) {
+                  toolCall.function.arguments += (_t = (_s = toolCallDelta.function) == null ? void 0 : _s.arguments) != null ? _t : "";
                 }
                 controller.enqueue({
                   type: "tool-call-delta",
                   toolCallType: "function",
                   toolCallId: toolCall.id,
                   toolName: toolCall.function.name,
-                  argsTextDelta: (_i = toolCallDelta.function.arguments) != null ? _i : ""
+                  argsTextDelta: (_u = toolCallDelta.function.arguments) != null ? _u : ""
                 });
-                if (((_j = toolCall.function) == null ? void 0 : _j.name) != null && ((_k = toolCall.function) == null ? void 0 : _k.arguments) != null && isParsableJson(toolCall.function.arguments)) {
+                if (((_v = toolCall.function) == null ? void 0 : _v.name) != null && ((_w = toolCall.function) == null ? void 0 : _w.arguments) != null && isParsableJson(toolCall.function.arguments)) {
                   controller.enqueue({
                     type: "tool-call",
                     toolCallType: "function",
-                    toolCallId: (_l = toolCall.id) != null ? _l : generateId(),
+                    toolCallId: (_x = toolCall.id) != null ? _x : generateId(),
                     toolName: toolCall.function.name,
                     args: toolCall.function.arguments
                   });
@@ -792,6 +784,7 @@ var OpenAIChatLanguageModel = class {
 var openaiTokenUsageSchema = z3.object({
   prompt_tokens: z3.number().nullish(),
   completion_tokens: z3.number().nullish(),
+  total_tokens: z3.number().nullish(),
   prompt_tokens_details: z3.object({
     cached_tokens: z3.number().nullish()
   }).nullish(),
@@ -863,6 +856,20 @@ var openaiChatChunkSchema = z3.union([
             })
           ).nullish()
         }).nullish(),
+        logprobs: z3.object({
+          content: z3.array(
+            z3.object({
+              token: z3.string(),
+              logprob: z3.number(),
+              top_logprobs: z3.array(
+                z3.object({
+                  token: z3.string(),
+                  logprob: z3.number()
+                })
+              )
+            })
+          ).nullish()
+        }).nullish(),
         finish_reason: z3.string().nullish(),
         index: z3.number()
       })
@@ -1133,6 +1140,7 @@ var OpenAICompletionLanguageModel = class {
     };
   }
   async doGenerate(options) {
+    var _a, _b, _c;
     const { args, warnings } = await this.getArgs(options);
     const {
       responseHeaders,
@@ -1160,8 +1168,9 @@ var OpenAICompletionLanguageModel = class {
     return {
       content: [{ type: "text", text: choice.text }],
       usage: {
-        inputTokens: response.usage.prompt_tokens,
-        outputTokens: response.usage.completion_tokens
+        inputTokens: (_a = response.usage) == null ? void 0 : _a.prompt_tokens,
+        outputTokens: (_b = response.usage) == null ? void 0 : _b.completion_tokens,
+        totalTokens: (_c = response.usage) == null ? void 0 : _c.total_tokens
       },
       finishReason: mapOpenAIFinishReason(choice.finish_reason),
       request: { body: args },
@@ -1179,8 +1188,9 @@ var OpenAICompletionLanguageModel = class {
     const body = {
       ...args,
       stream: true,
-      // only include stream_options when in strict compatibility mode:
-      stream_options: this.config.compatibility === "strict" ? { include_usage: true } : void 0
+      stream_options: {
+        include_usage: true
+      }
     };
     const { responseHeaders, value: response } = await postJsonToApi2({
       url: this.config.url({
@@ -1200,7 +1210,8 @@ var OpenAICompletionLanguageModel = class {
     const providerMetadata = { openai: {} };
     const usage = {
       inputTokens: void 0,
-      outputTokens: void 0
+      outputTokens: void 0,
+      totalTokens: void 0
     };
     let isFirstChunk = true;
     return {
@@ -1231,6 +1242,7 @@ var OpenAICompletionLanguageModel = class {
             if (value.usage != null) {
               usage.inputTokens = value.usage.prompt_tokens;
               usage.outputTokens = value.usage.completion_tokens;
+              usage.totalTokens = value.usage.total_tokens;
             }
             const choice = value.choices[0];
             if ((choice == null ? void 0 : choice.finish_reason) != null) {
@@ -1261,6 +1273,11 @@ var OpenAICompletionLanguageModel = class {
     };
   }
 };
+var usageSchema = z5.object({
+  prompt_tokens: z5.number(),
+  completion_tokens: z5.number(),
+  total_tokens: z5.number()
+});
 var openaiCompletionResponseSchema = z5.object({
   id: z5.string().nullish(),
   created: z5.number().nullish(),
@@ -1276,10 +1293,7 @@ var openaiCompletionResponseSchema = z5.object({
       }).nullish()
     })
   ),
-  usage: z5.object({
-    prompt_tokens: z5.number(),
-    completion_tokens: z5.number()
-  })
+  usage: usageSchema.nullish()
 });
 var openaiCompletionChunkSchema = z5.union([
   z5.object({
@@ -1298,10 +1312,7 @@ var openaiCompletionChunkSchema = z5.union([
         }).nullish()
       })
     ),
-    usage: z5.object({
-      prompt_tokens: z5.number(),
-      completion_tokens: z5.number()
-    }).nullish()
+    usage: usageSchema.nullish()
   }),
   openaiErrorDataSchema
 ]);
@@ -1518,25 +1529,25 @@ var openAITranscriptionProviderOptions = z9.object({
   /**
    * Additional information to include in the transcription response.
    */
-  include: z9.array(z9.string()).nullish(),
+  include: z9.array(z9.string()).optional(),
   /**
    * The language of the input audio in ISO-639-1 format.
    */
-  language: z9.string().nullish(),
+  language: z9.string().optional(),
   /**
    * An optional text to guide the model's style or continue a previous audio segment.
    */
-  prompt: z9.string().nullish(),
+  prompt: z9.string().optional(),
   /**
    * The sampling temperature, between 0 and 1.
    * @default 0
    */
-  temperature: z9.number().min(0).max(1).default(0).nullish(),
+  temperature: z9.number().min(0).max(1).default(0).optional(),
   /**
    * The timestamp granularities to populate for this transcription.
    * @default ['segment']
    */
-  timestampGranularities: z9.array(z9.enum(["word", "segment"])).default(["segment"]).nullish()
+  timestampGranularities: z9.array(z9.enum(["word", "segment"])).default(["segment"]).optional()
 });
 
 // src/openai-transcription-model.ts
@@ -2222,7 +2233,7 @@ var OpenAIResponsesLanguageModel = class {
             ])
           ),
           incomplete_details: z12.object({ reason: z12.string() }).nullable(),
-          usage: usageSchema
+          usage: usageSchema2
         })
       ),
       abortSignal: options.abortSignal,
@@ -2276,7 +2287,10 @@ var OpenAIResponsesLanguageModel = class {
       }),
       usage: {
         inputTokens: response.usage.input_tokens,
-        outputTokens: response.usage.output_tokens
+        outputTokens: response.usage.output_tokens,
+        totalTokens: response.usage.input_tokens + response.usage.output_tokens,
+        reasoningTokens: (_f = (_e = response.usage.output_tokens_details) == null ? void 0 : _e.reasoning_tokens) != null ? _f : void 0,
+        cachedInputTokens: (_h = (_g = response.usage.input_tokens_details) == null ? void 0 : _g.cached_tokens) != null ? _h : void 0
       },
       request: { body },
       response: {
@@ -2288,9 +2302,7 @@ var OpenAIResponsesLanguageModel = class {
       },
       providerMetadata: {
         openai: {
-          responseId: response.id,
-          cachedPromptTokens: (_f = (_e = response.usage.input_tokens_details) == null ? void 0 : _e.cached_tokens) != null ? _f : null,
-          reasoningTokens: (_h = (_g = response.usage.output_tokens_details) == null ? void 0 : _g.reasoning_tokens) != null ? _h : null
+          responseId: response.id
         }
       },
       warnings
@@ -2319,10 +2331,9 @@ var OpenAIResponsesLanguageModel = class {
     let finishReason = "unknown";
     const usage = {
       inputTokens: void 0,
-      outputTokens: void 0
+      outputTokens: void 0,
+      totalTokens: void 0
     };
-    let cachedPromptTokens = null;
-    let reasoningTokens = null;
     let responseId = null;
     const ongoingToolCalls = {};
     let hasToolCalls = false;
@@ -2400,8 +2411,9 @@ var OpenAIResponsesLanguageModel = class {
               });
               usage.inputTokens = value.response.usage.input_tokens;
               usage.outputTokens = value.response.usage.output_tokens;
-              cachedPromptTokens = (_c = (_b = value.response.usage.input_tokens_details) == null ? void 0 : _b.cached_tokens) != null ? _c : cachedPromptTokens;
-              reasoningTokens = (_e = (_d = value.response.usage.output_tokens_details) == null ? void 0 : _d.reasoning_tokens) != null ? _e : reasoningTokens;
+              usage.totalTokens = value.response.usage.input_tokens + value.response.usage.output_tokens;
+              usage.reasoningTokens = (_c = (_b = value.response.usage.output_tokens_details) == null ? void 0 : _b.reasoning_tokens) != null ? _c : void 0;
+              usage.cachedInputTokens = (_e = (_d = value.response.usage.input_tokens_details) == null ? void 0 : _d.cached_tokens) != null ? _e : void 0;
             } else if (isResponseAnnotationAddedChunk(value)) {
               controller.enqueue({
                 type: "source",
@@ -2417,13 +2429,9 @@ var OpenAIResponsesLanguageModel = class {
               type: "finish",
               finishReason,
               usage,
-              ...(cachedPromptTokens != null || reasoningTokens != null) && {
-                providerMetadata: {
-                  openai: {
-                    responseId,
-                    cachedPromptTokens,
-                    reasoningTokens
-                  }
+              providerMetadata: {
+                openai: {
+                  responseId
                 }
               }
             });
@@ -2435,7 +2443,7 @@ var OpenAIResponsesLanguageModel = class {
     };
   }
 };
-var usageSchema = z12.object({
+var usageSchema2 = z12.object({
   input_tokens: z12.number(),
   input_tokens_details: z12.object({ cached_tokens: z12.number().nullish() }).nullish(),
   output_tokens: z12.number(),
@@ -2449,7 +2457,7 @@ var responseFinishedChunkSchema = z12.object({
   type: z12.enum(["response.completed", "response.incomplete"]),
   response: z12.object({
     incomplete_details: z12.object({ reason: z12.string() }).nullish(),
-    usage: usageSchema
+    usage: usageSchema2
   })
 });
 var responseCreatedChunkSchema = z12.object({