@ai-sdk/openai 2.0.20 → 2.0.22

package/dist/internal/index.mjs

@@ -424,7 +424,7 @@ var webSearchPreview = createProviderDefinedToolFactory2({
     action: z4.discriminatedUnion("type", [
       z4.object({
         type: z4.literal("search"),
-        query: z4.string()
+        query: z4.string().nullish()
       }),
       z4.object({
         type: z4.literal("open_page"),
@@ -1938,8 +1938,12 @@ var OpenAITranscriptionModel = class {
         include: openAIOptions.include,
         language: openAIOptions.language,
         prompt: openAIOptions.prompt,
-        response_format: "verbose_json",
-        // always use verbose_json to get segments
+        // https://platform.openai.com/docs/api-reference/audio/createTranscription#audio_createtranscription-response_format
+        // prefer verbose_json to get segments for models that support it
+        response_format: [
+          "gpt-4o-transcribe",
+          "gpt-4o-mini-transcribe"
+        ].includes(this.modelId) ? "json" : "verbose_json",
         temperature: openAIOptions.temperature,
         timestamp_granularities: openAIOptions.timestampGranularities
       };
@@ -2490,7 +2494,7 @@ var webSearchCallItem = z16.object({
   action: z16.discriminatedUnion("type", [
     z16.object({
       type: z16.literal("search"),
-      query: z16.string()
+      query: z16.string().nullish()
     }),
     z16.object({
       type: z16.literal("open_page"),
@@ -2699,7 +2703,7 @@ var OpenAIResponsesLanguageModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n;
+    var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q;
     const { args: body, warnings } = await this.getArgs(options);
     const url = this.config.url({
       path: "/responses",
@@ -2745,10 +2749,12 @@ var OpenAIResponsesLanguageModel = class {
                         }),
                         z16.object({
                           type: z16.literal("file_citation"),
-                          start_index: z16.number(),
-                          end_index: z16.number(),
                           file_id: z16.string(),
-                          quote: z16.string()
+                          filename: z16.string().nullish(),
+                          index: z16.number().nullish(),
+                          start_index: z16.number().nullish(),
+                          end_index: z16.number().nullish(),
+                          quote: z16.string().nullish()
                         })
                       ])
                     )
@@ -2866,8 +2872,8 @@ var OpenAIResponsesLanguageModel = class {
                   sourceType: "document",
                   id: (_i = (_h = (_g = this.config).generateId) == null ? void 0 : _h.call(_g)) != null ? _i : generateId2(),
                   mediaType: "text/plain",
-                  title: annotation.quote,
-                  filename: annotation.file_id
+                  title: (_k = (_j = annotation.quote) != null ? _j : annotation.filename) != null ? _k : "Document",
+                  filename: (_l = annotation.filename) != null ? _l : annotation.file_id
                 });
               }
             }
@@ -2958,15 +2964,15 @@ var OpenAIResponsesLanguageModel = class {
     return {
       content,
       finishReason: mapOpenAIResponseFinishReason({
-        finishReason: (_j = response.incomplete_details) == null ? void 0 : _j.reason,
+        finishReason: (_m = response.incomplete_details) == null ? void 0 : _m.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: (_l = (_k = response.usage.output_tokens_details) == null ? void 0 : _k.reasoning_tokens) != null ? _l : void 0,
-        cachedInputTokens: (_n = (_m = response.usage.input_tokens_details) == null ? void 0 : _m.cached_tokens) != null ? _n : void 0
+        reasoningTokens: (_o = (_n = response.usage.output_tokens_details) == null ? void 0 : _n.reasoning_tokens) != null ? _o : void 0,
+        cachedInputTokens: (_q = (_p = response.usage.input_tokens_details) == null ? void 0 : _p.cached_tokens) != null ? _q : void 0
       },
       request: { body },
       response: {
@@ -3018,7 +3024,7 @@ var OpenAIResponsesLanguageModel = class {
             controller.enqueue({ type: "stream-start", warnings });
           },
           transform(chunk, controller) {
-            var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r;
+            var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u;
             if (options.includeRawChunks) {
               controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
             }
@@ -3284,8 +3290,8 @@ var OpenAIResponsesLanguageModel = class {
                   sourceType: "document",
                   id: (_r = (_q = (_p = self.config).generateId) == null ? void 0 : _q.call(_p)) != null ? _r : generateId2(),
                   mediaType: "text/plain",
-                  title: value.annotation.quote,
-                  filename: value.annotation.file_id
+                  title: (_t = (_s = value.annotation.quote) != null ? _s : value.annotation.filename) != null ? _t : "Document",
+                  filename: (_u = value.annotation.filename) != null ? _u : value.annotation.file_id
                 });
               }
             } else if (isErrorChunk(value)) {
@@ -3463,7 +3469,11 @@ var responseAnnotationAddedSchema = z16.object({
     z16.object({
       type: z16.literal("file_citation"),
       file_id: z16.string(),
-      quote: z16.string()
+      filename: z16.string().nullish(),
+      index: z16.number().nullish(),
+      start_index: z16.number().nullish(),
+      end_index: z16.number().nullish(),
+      quote: z16.string().nullish()
     })
   ])
 });