@ai-sdk/openai 2.0.19 → 2.0.21

package/CHANGELOG.md

@@ -1,5 +1,24 @@
 # @ai-sdk/openai
 
+## 2.0.21
+
+### Patch Changes
+
+- d18856a: fix(provider/openai): support websearch tool results without query property
+- 15271d6: fix(provider/openai): do not set `response_format` to `verbose_json` if model is `gpt-4o-transcribe` or `gpt-4o-mini-transcribe`
+
+  These two models do not support it:
+  https://platform.openai.com/docs/api-reference/audio/createTranscription#audio_createtranscription-response_format
+
+- Updated dependencies [1b5a3d3]
+  - @ai-sdk/provider-utils@3.0.6
+
+## 2.0.20
+
+### Patch Changes
+
+- 974de40: fix(provider/ai): do not set `.providerMetadata.openai.logprobs` to an array of empty arrays when using `streamText()`
+
 ## 2.0.19
 
 ### Patch Changes

package/dist/index.js

@@ -440,7 +440,7 @@ var webSearchPreview = (0, import_provider_utils4.createProviderDefinedToolFacto
     action: import_v44.z.discriminatedUnion("type", [
       import_v44.z.object({
         type: import_v44.z.literal("search"),
-        query: import_v44.z.string()
+        query: import_v44.z.string().nullish()
       }),
       import_v44.z.object({
         type: import_v44.z.literal("open_page"),
@@ -2151,7 +2151,7 @@ var webSearchCallItem = import_v413.z.object({
   action: import_v413.z.discriminatedUnion("type", [
     import_v413.z.object({
       type: import_v413.z.literal("search"),
-      query: import_v413.z.string()
+      query: import_v413.z.string().nullish()
     }),
     import_v413.z.object({
       type: import_v413.z.literal("open_page"),
@@ -2679,7 +2679,7 @@ 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;
+            var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r;
             if (options.includeRawChunks) {
               controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
             }
@@ -2890,12 +2890,12 @@ var OpenAIResponsesLanguageModel = class {
                 id: value.item_id,
                 delta: value.delta
               });
-              if (value.logprobs) {
+              if (((_d = (_c = options.providerOptions) == null ? void 0 : _c.openai) == null ? void 0 : _d.logprobs) && 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(
+                (_e = activeReasoning[value.item_id]) == null ? void 0 : _e.summaryParts.push(
                   value.summary_index
                 );
                 controller.enqueue({
@@ -2904,7 +2904,7 @@ var OpenAIResponsesLanguageModel = class {
                   providerMetadata: {
                     openai: {
                       itemId: value.item_id,
-                      reasoningEncryptedContent: (_e = (_d = activeReasoning[value.item_id]) == null ? void 0 : _d.encryptedContent) != null ? _e : null
+                      reasoningEncryptedContent: (_g = (_f = activeReasoning[value.item_id]) == null ? void 0 : _f.encryptedContent) != null ? _g : null
                     }
                   }
                 });
@@ -2922,20 +2922,20 @@ var OpenAIResponsesLanguageModel = class {
               });
             } else if (isResponseFinishedChunk(value)) {
               finishReason = mapOpenAIResponseFinishReason({
-                finishReason: (_f = value.response.incomplete_details) == null ? void 0 : _f.reason,
+                finishReason: (_h = value.response.incomplete_details) == null ? void 0 : _h.reason,
                 hasToolCalls
               });
               usage.inputTokens = value.response.usage.input_tokens;
               usage.outputTokens = value.response.usage.output_tokens;
               usage.totalTokens = value.response.usage.input_tokens + value.response.usage.output_tokens;
-              usage.reasoningTokens = (_h = (_g = value.response.usage.output_tokens_details) == null ? void 0 : _g.reasoning_tokens) != null ? _h : void 0;
-              usage.cachedInputTokens = (_j = (_i = value.response.usage.input_tokens_details) == null ? void 0 : _i.cached_tokens) != null ? _j : void 0;
+              usage.reasoningTokens = (_j = (_i = value.response.usage.output_tokens_details) == null ? void 0 : _i.reasoning_tokens) != null ? _j : void 0;
+              usage.cachedInputTokens = (_l = (_k = value.response.usage.input_tokens_details) == null ? void 0 : _k.cached_tokens) != null ? _l : void 0;
             } else if (isResponseAnnotationAddedChunk(value)) {
               if (value.annotation.type === "url_citation") {
                 controller.enqueue({
                   type: "source",
                   sourceType: "url",
-                  id: (_m = (_l = (_k = self.config).generateId) == null ? void 0 : _l.call(_k)) != null ? _m : (0, import_provider_utils12.generateId)(),
+                  id: (_o = (_n = (_m = self.config).generateId) == null ? void 0 : _n.call(_m)) != null ? _o : (0, import_provider_utils12.generateId)(),
                   url: value.annotation.url,
                   title: value.annotation.title
                 });
@@ -2943,7 +2943,7 @@ var OpenAIResponsesLanguageModel = class {
                 controller.enqueue({
                   type: "source",
                   sourceType: "document",
-                  id: (_p = (_o = (_n = self.config).generateId) == null ? void 0 : _o.call(_n)) != null ? _p : (0, import_provider_utils12.generateId)(),
+                  id: (_r = (_q = (_p = self.config).generateId) == null ? void 0 : _q.call(_p)) != null ? _r : (0, import_provider_utils12.generateId)(),
                   mediaType: "text/plain",
                   title: value.annotation.quote,
                   filename: value.annotation.file_id
@@ -3486,8 +3486,12 @@ 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
       };