@ai-sdk/openai 2.0.11 → 2.0.13

package/dist/index.mjs

@@ -1125,13 +1125,13 @@ var openaiChatChunkSchema = z5.union([
   openaiErrorDataSchema
 ]);
 function isReasoningModel(modelId) {
-  return modelId.startsWith("o") || modelId.startsWith("gpt-5");
+  return (modelId.startsWith("o") || modelId.startsWith("gpt-5")) && !modelId.startsWith("gpt-5-chat");
 }
 function supportsFlexProcessing(modelId) {
-  return modelId.startsWith("o3") || modelId.startsWith("o4-mini") || modelId.startsWith("gpt-5");
+  return modelId.startsWith("o3") || modelId.startsWith("o4-mini") || modelId.startsWith("gpt-5") && !modelId.startsWith("gpt-5-chat");
 }
 function supportsPriorityProcessing(modelId) {
-  return modelId.startsWith("gpt-4") || modelId.startsWith("gpt-5-mini") || modelId.startsWith("gpt-5") && !modelId.startsWith("gpt-5-nano") || modelId.startsWith("o3") || modelId.startsWith("o4-mini");
+  return modelId.startsWith("gpt-4") || modelId.startsWith("gpt-5-mini") || modelId.startsWith("gpt-5") && !modelId.startsWith("gpt-5-nano") && !modelId.startsWith("gpt-5-chat") || modelId.startsWith("o3") || modelId.startsWith("o4-mini");
 }
 function getSystemMessageMode(modelId) {
   var _a, _b;
@@ -2150,6 +2150,19 @@ function prepareResponsesTools({
 }
 
 // src/responses/openai-responses-language-model.ts
+var TOP_LOGPROBS_MAX = 20;
+var LOGPROBS_SCHEMA = z13.array(
+  z13.object({
+    token: z13.string(),
+    logprob: z13.number(),
+    top_logprobs: z13.array(
+      z13.object({
+        token: z13.string(),
+        logprob: z13.number()
+      })
+    )
+  })
+);
 var OpenAIResponsesLanguageModel = class {
   constructor(modelId, config) {
     this.specificationVersion = "v2";
@@ -2213,6 +2226,8 @@ var OpenAIResponsesLanguageModel = class {
       schema: openaiResponsesProviderOptionsSchema
     });
     const strictJsonSchema = (_a = openaiOptions == null ? void 0 : openaiOptions.strictJsonSchema) != null ? _a : false;
+    const topLogprobs = typeof (openaiOptions == null ? void 0 : openaiOptions.logprobs) === "number" ? openaiOptions == null ? void 0 : openaiOptions.logprobs : (openaiOptions == null ? void 0 : openaiOptions.logprobs) === true ? TOP_LOGPROBS_MAX : void 0;
+    const openaiOptionsInclude = topLogprobs ? Array.isArray(openaiOptions == null ? void 0 : openaiOptions.include) ? [...openaiOptions == null ? void 0 : openaiOptions.include, "message.output_text.logprobs"] : ["message.output_text.logprobs"] : openaiOptions == null ? void 0 : openaiOptions.include;
     const baseArgs = {
       model: this.modelId,
       input: messages,
@@ -2243,9 +2258,10 @@ var OpenAIResponsesLanguageModel = class {
       user: openaiOptions == null ? void 0 : openaiOptions.user,
       instructions: openaiOptions == null ? void 0 : openaiOptions.instructions,
       service_tier: openaiOptions == null ? void 0 : openaiOptions.serviceTier,
-      include: openaiOptions == null ? void 0 : openaiOptions.include,
+      include: openaiOptionsInclude,
       prompt_cache_key: openaiOptions == null ? void 0 : openaiOptions.promptCacheKey,
       safety_identifier: openaiOptions == null ? void 0 : openaiOptions.safetyIdentifier,
+      top_logprobs: topLogprobs,
       // model-specific settings:
       ...modelConfig.isReasoningModel && ((openaiOptions == null ? void 0 : openaiOptions.reasoningEffort) != null || (openaiOptions == null ? void 0 : openaiOptions.reasoningSummary) != null) && {
         reasoning: {
@@ -2329,7 +2345,7 @@ var OpenAIResponsesLanguageModel = 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;
     const { args: body, warnings } = await this.getArgs(options);
     const url = this.config.url({
       path: "/responses",
@@ -2363,6 +2379,7 @@ var OpenAIResponsesLanguageModel = class {
                   z13.object({
                     type: z13.literal("output_text"),
                     text: z13.string(),
+                    logprobs: LOGPROBS_SCHEMA.nullish(),
                     annotations: z13.array(
                       z13.object({
                         type: z13.literal("url_citation"),
@@ -2440,6 +2457,7 @@ var OpenAIResponsesLanguageModel = class {
       });
     }
     const content = [];
+    const logprobs = [];
     for (const part of response.output) {
       switch (part.type) {
         case "reasoning": {
@@ -2462,6 +2480,9 @@ var OpenAIResponsesLanguageModel = class {
         }
         case "message": {
           for (const contentPart of part.content) {
+            if (((_c = (_b = options.providerOptions) == null ? void 0 : _b.openai) == null ? void 0 : _c.logprobs) && contentPart.logprobs) {
+              logprobs.push(contentPart.logprobs);
+            }
             content.push({
               type: "text",
               text: contentPart.text,
@@ -2475,7 +2496,7 @@ var OpenAIResponsesLanguageModel = class {
               content.push({
                 type: "source",
                 sourceType: "url",
-                id: (_d = (_c = (_b = this.config).generateId) == null ? void 0 : _c.call(_b)) != null ? _d : generateId2(),
+                id: (_f = (_e = (_d = this.config).generateId) == null ? void 0 : _e.call(_d)) != null ? _f : generateId2(),
                 url: annotation.url,
                 title: annotation.title
               });
@@ -2558,18 +2579,24 @@ var OpenAIResponsesLanguageModel = class {
         }
       }
     }
+    const providerMetadata = {
+      openai: { responseId: response.id }
+    };
+    if (logprobs.length > 0) {
+      providerMetadata.openai.logprobs = logprobs;
+    }
     return {
       content,
       finishReason: mapOpenAIResponseFinishReason({
-        finishReason: (_e = response.incomplete_details) == null ? void 0 : _e.reason,
+        finishReason: (_g = response.incomplete_details) == null ? void 0 : _g.reason,
         hasToolCalls: content.some((part) => part.type === "tool-call")
       }),
       usage: {
         inputTokens: response.usage.input_tokens,
         outputTokens: response.usage.output_tokens,
         totalTokens: response.usage.input_tokens + response.usage.output_tokens,
-        reasoningTokens: (_g = (_f = response.usage.output_tokens_details) == null ? void 0 : _f.reasoning_tokens) != null ? _g : void 0,
-        cachedInputTokens: (_i = (_h = response.usage.input_tokens_details) == null ? void 0 : _h.cached_tokens) != null ? _i : void 0
+        reasoningTokens: (_i = (_h = response.usage.output_tokens_details) == null ? void 0 : _h.reasoning_tokens) != null ? _i : void 0,
+        cachedInputTokens: (_k = (_j = response.usage.input_tokens_details) == null ? void 0 : _j.cached_tokens) != null ? _k : void 0
       },
       request: { body },
       response: {
@@ -2579,11 +2606,7 @@ var OpenAIResponsesLanguageModel = class {
         headers: responseHeaders,
         body: rawResponse
       },
-      providerMetadata: {
-        openai: {
-          responseId: response.id
-        }
-      },
+      providerMetadata,
       warnings
     };
   }
@@ -2613,6 +2636,7 @@ var OpenAIResponsesLanguageModel = class {
       outputTokens: void 0,
       totalTokens: void 0
     };
+    const logprobs = [];
     let responseId = null;
     const ongoingToolCalls = {};
     let hasToolCalls = false;
@@ -2838,6 +2862,9 @@ var OpenAIResponsesLanguageModel = class {
                 id: value.item_id,
                 delta: value.delta
               });
+              if (value.logprobs) {
+                logprobs.push(value.logprobs);
+              }
             } else if (isResponseReasoningSummaryPartAddedChunk(value)) {
               if (value.summary_index > 0) {
                 (_c = activeReasoning[value.item_id]) == null ? void 0 : _c.summaryParts.push(
@@ -2888,15 +2915,19 @@ var OpenAIResponsesLanguageModel = class {
             }
           },
           flush(controller) {
+            const providerMetadata = {
+              openai: {
+                responseId
+              }
+            };
+            if (logprobs.length > 0) {
+              providerMetadata.openai.logprobs = logprobs;
+            }
             controller.enqueue({
               type: "finish",
               finishReason,
               usage,
-              providerMetadata: {
-                openai: {
-                  responseId
-                }
-              }
+              providerMetadata
             });
           }
         })
@@ -2915,7 +2946,8 @@ var usageSchema2 = z13.object({
 var textDeltaChunkSchema = z13.object({
   type: z13.literal("response.output_text.delta"),
   item_id: z13.string(),
-  delta: z13.string()
+  delta: z13.string(),
+  logprobs: LOGPROBS_SCHEMA.nullish()
 });
 var errorChunkSchema = z13.object({
   type: z13.literal("error"),
@@ -3112,6 +3144,13 @@ function isErrorChunk(chunk) {
   return chunk.type === "error";
 }
 function getResponsesModelConfig(modelId) {
+  if (modelId.startsWith("gpt-5-chat")) {
+    return {
+      isReasoningModel: false,
+      systemMessageMode: "system",
+      requiredAutoTruncation: false
+    };
+  }
   if (modelId.startsWith("o") || modelId.startsWith("gpt-5") || modelId.startsWith("codex-") || modelId.startsWith("computer-use")) {
     if (modelId.startsWith("o1-mini") || modelId.startsWith("o1-preview")) {
       return {
@@ -3133,10 +3172,10 @@ function getResponsesModelConfig(modelId) {
   };
 }
 function supportsFlexProcessing2(modelId) {
-  return modelId.startsWith("o3") || modelId.startsWith("o4-mini") || modelId.startsWith("gpt-5");
+  return modelId.startsWith("o3") || modelId.startsWith("o4-mini") || modelId.startsWith("gpt-5") && !modelId.startsWith("gpt-5-chat");
 }
 function supportsPriorityProcessing2(modelId) {
-  return modelId.startsWith("gpt-4") || modelId.startsWith("gpt-5-mini") || modelId.startsWith("gpt-5") && !modelId.startsWith("gpt-5-nano") || modelId.startsWith("o3") || modelId.startsWith("o4-mini");
+  return modelId.startsWith("gpt-4") || modelId.startsWith("gpt-5-mini") || modelId.startsWith("gpt-5") && !modelId.startsWith("gpt-5-nano") && !modelId.startsWith("gpt-5-chat") || modelId.startsWith("o3") || modelId.startsWith("o4-mini");
 }
 var openaiResponsesProviderOptionsSchema = z13.object({
   metadata: z13.any().nullish(),
@@ -3149,10 +3188,29 @@ var openaiResponsesProviderOptionsSchema = z13.object({
   instructions: z13.string().nullish(),
   reasoningSummary: z13.string().nullish(),
   serviceTier: z13.enum(["auto", "flex", "priority"]).nullish(),
-  include: z13.array(z13.enum(["reasoning.encrypted_content", "file_search_call.results"])).nullish(),
+  include: z13.array(
+    z13.enum([
+      "reasoning.encrypted_content",
+      "file_search_call.results",
+      "message.output_text.logprobs"
+    ])
+  ).nullish(),
   textVerbosity: z13.enum(["low", "medium", "high"]).nullish(),
   promptCacheKey: z13.string().nullish(),
-  safetyIdentifier: z13.string().nullish()
+  safetyIdentifier: z13.string().nullish(),
+  /**
+   * Return the log probabilities of the tokens.
+   *
+   * Setting to true will return the log probabilities of the tokens that
+   * were generated.
+   *
+   * Setting to a number will return the log probabilities of the top n
+   * tokens that were generated.
+   *
+   * @see https://platform.openai.com/docs/api-reference/responses/create
+   * @see https://cookbook.openai.com/examples/using_logprobs
+   */
+  logprobs: z13.union([z13.boolean(), z13.number().min(1).max(TOP_LOGPROBS_MAX)]).optional()
 });
 
 // src/speech/openai-speech-model.ts
@@ -3393,6 +3451,8 @@ var OpenAITranscriptionModel = class {
         include: openAIOptions.include,
         language: openAIOptions.language,
         prompt: openAIOptions.prompt,
+        response_format: "verbose_json",
+        // always use verbose_json to get segments
         temperature: openAIOptions.temperature,
         timestamp_granularities: openAIOptions.timestampGranularities
       };
@@ -3408,7 +3468,7 @@ var OpenAITranscriptionModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e, _f;
+    var _a, _b, _c, _d, _e, _f, _g, _h;
     const currentDate = (_c = (_b = (_a = this.config._internal) == null ? void 0 : _a.currentDate) == null ? void 0 : _b.call(_a)) != null ? _c : /* @__PURE__ */ new Date();
     const { formData, warnings } = await this.getArgs(options);
     const {
@@ -3432,13 +3492,17 @@ var OpenAITranscriptionModel = class {
     const language = response.language != null && response.language in languageMap ? languageMap[response.language] : void 0;
     return {
       text: response.text,
-      segments: (_e = (_d = response.words) == null ? void 0 : _d.map((word) => ({
+      segments: (_g = (_f = (_d = response.segments) == null ? void 0 : _d.map((segment) => ({
+        text: segment.text,
+        startSecond: segment.start,
+        endSecond: segment.end
+      }))) != null ? _f : (_e = response.words) == null ? void 0 : _e.map((word) => ({
         text: word.word,
         startSecond: word.start,
         endSecond: word.end
-      }))) != null ? _e : [],
+      }))) != null ? _g : [],
       language,
-      durationInSeconds: (_f = response.duration) != null ? _f : void 0,
+      durationInSeconds: (_h = response.duration) != null ? _h : void 0,
       warnings,
       response: {
         timestamp: currentDate,
@@ -3459,6 +3523,20 @@ var openaiTranscriptionResponseSchema = z16.object({
       start: z16.number(),
       end: z16.number()
     })
+  ).nullish(),
+  segments: z16.array(
+    z16.object({
+      id: z16.number(),
+      seek: z16.number(),
+      start: z16.number(),
+      end: z16.number(),
+      text: z16.string(),
+      tokens: z16.array(z16.number()),
+      temperature: z16.number(),
+      avg_logprob: z16.number(),
+      compression_ratio: z16.number(),
+      no_speech_prob: z16.number()
+    })
   ).nullish()
 });