@ai-sdk/openai 2.0.10 → 2.0.12

package/dist/internal/index.mjs

@@ -316,7 +316,20 @@ var openaiProviderOptions = z2.object({
    * Controls the verbosity of the model's responses.
    * Lower values will result in more concise responses, while higher values will result in more verbose responses.
    */
-  textVerbosity: z2.enum(["low", "medium", "high"]).optional()
+  textVerbosity: z2.enum(["low", "medium", "high"]).optional(),
+  /**
+   * A cache key for prompt caching. Allows manual control over prompt caching behavior.
+   * Useful for improving cache hit rates and working around automatic caching issues.
+   */
+  promptCacheKey: z2.string().optional(),
+  /**
+   * A stable identifier used to help detect users of your application
+   * that may be violating OpenAI's usage policies. The IDs should be a
+   * string that uniquely identifies each user. We recommend hashing their
+   * username or email address, in order to avoid sending us any identifying
+   * information.
+   */
+  safetyIdentifier: z2.string().optional()
 });
 
 // src/chat/openai-chat-prepare-tools.ts
@@ -589,6 +602,8 @@ var OpenAIChatLanguageModel = class {
       prediction: openaiOptions.prediction,
       reasoning_effort: openaiOptions.reasoningEffort,
       service_tier: openaiOptions.serviceTier,
+      prompt_cache_key: openaiOptions.promptCacheKey,
+      safety_identifier: openaiOptions.safetyIdentifier,
       // messages:
       messages
     };
@@ -1104,13 +1119,13 @@ 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;
@@ -1907,6 +1922,8 @@ 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
       };
@@ -1922,7 +1939,7 @@ 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 {
@@ -1946,13 +1963,17 @@ 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,
@@ -1973,6 +1994,20 @@ var openaiTranscriptionResponseSchema = z12.object({
       start: z12.number(),
       end: z12.number()
     })
+  ).nullish(),
+  segments: z12.array(
+    z12.object({
+      id: z12.number(),
+      seek: z12.number(),
+      start: z12.number(),
+      end: z12.number(),
+      text: z12.string(),
+      tokens: z12.array(z12.number()),
+      temperature: z12.number(),
+      avg_logprob: z12.number(),
+      compression_ratio: z12.number(),
+      no_speech_prob: z12.number()
+    })
   ).nullish()
 });
 
@@ -2109,9 +2144,14 @@ import {
 import { parseProviderOptions as parseProviderOptions6 } from "@ai-sdk/provider-utils";
 import { z as z14 } from "zod/v4";
 import { convertToBase64 as convertToBase642 } from "@ai-sdk/provider-utils";
+function isFileId(data, prefixes) {
+  if (!prefixes) return false;
+  return prefixes.some((prefix) => data.startsWith(prefix));
+}
 async function convertToOpenAIResponsesMessages({
   prompt,
-  systemMessageMode
+  systemMessageMode,
+  fileIdPrefixes
 }) {
   var _a, _b, _c, _d, _e, _f;
   const messages = [];
@@ -2158,7 +2198,7 @@ async function convertToOpenAIResponsesMessages({
                   const mediaType = part.mediaType === "image/*" ? "image/jpeg" : part.mediaType;
                   return {
                     type: "input_image",
-                    ...part.data instanceof URL ? { image_url: part.data.toString() } : typeof part.data === "string" && part.data.startsWith("file-") ? { file_id: part.data } : {
+                    ...part.data instanceof URL ? { image_url: part.data.toString() } : typeof part.data === "string" && isFileId(part.data, fileIdPrefixes) ? { file_id: part.data } : {
                       image_url: `data:${mediaType};base64,${convertToBase642(part.data)}`
                     },
                     detail: (_b2 = (_a2 = part.providerOptions) == null ? void 0 : _a2.openai) == null ? void 0 : _b2.imageDetail
@@ -2171,7 +2211,7 @@ async function convertToOpenAIResponsesMessages({
                   }
                   return {
                     type: "input_file",
-                    ...typeof part.data === "string" && part.data.startsWith("file-") ? { file_id: part.data } : {
+                    ...typeof part.data === "string" && isFileId(part.data, fileIdPrefixes) ? { file_id: part.data } : {
                       filename: (_c2 = part.filename) != null ? _c2 : `part-${index}.pdf`,
                       file_data: `data:application/pdf;base64,${convertToBase642(part.data)}`
                     }
@@ -2479,7 +2519,8 @@ var OpenAIResponsesLanguageModel = class {
     }
     const { messages, warnings: messageWarnings } = await convertToOpenAIResponsesMessages({
       prompt,
-      systemMessageMode: modelConfig.systemMessageMode
+      systemMessageMode: modelConfig.systemMessageMode,
+      fileIdPrefixes: this.config.fileIdPrefixes
     });
     warnings.push(...messageWarnings);
     const openaiOptions = await parseProviderOptions7({
@@ -2519,6 +2560,8 @@ var OpenAIResponsesLanguageModel = class {
       instructions: openaiOptions == null ? void 0 : openaiOptions.instructions,
       service_tier: openaiOptions == null ? void 0 : openaiOptions.serviceTier,
       include: openaiOptions == null ? void 0 : openaiOptions.include,
+      prompt_cache_key: openaiOptions == null ? void 0 : openaiOptions.promptCacheKey,
+      safety_identifier: openaiOptions == null ? void 0 : openaiOptions.safetyIdentifier,
       // model-specific settings:
       ...modelConfig.isReasoningModel && ((openaiOptions == null ? void 0 : openaiOptions.reasoningEffort) != null || (openaiOptions == null ? void 0 : openaiOptions.reasoningSummary) != null) && {
         reasoning: {
@@ -2668,7 +2711,18 @@ var OpenAIResponsesLanguageModel = class {
               z16.object({
                 type: z16.literal("file_search_call"),
                 id: z16.string(),
-                status: z16.string().optional()
+                status: z16.string().optional(),
+                queries: z16.array(z16.string()).nullish(),
+                results: z16.array(
+                  z16.object({
+                    attributes: z16.object({
+                      file_id: z16.string(),
+                      filename: z16.string(),
+                      score: z16.number(),
+                      text: z16.string()
+                    })
+                  })
+                ).nullish()
               }),
               z16.object({
                 type: z16.literal("reasoning"),
@@ -2810,7 +2864,9 @@ var OpenAIResponsesLanguageModel = class {
             toolName: "file_search",
             result: {
               type: "file_search_tool_result",
-              status: part.status || "completed"
+              status: part.status || "completed",
+              ...part.queries && { queries: part.queries },
+              ...part.results && { results: part.results }
             },
             providerExecuted: true
           });
@@ -2925,6 +2981,16 @@ var OpenAIResponsesLanguageModel = class {
                   id: value.item.id,
                   toolName: "computer_use"
                 });
+              } else if (value.item.type === "file_search_call") {
+                ongoingToolCalls[value.output_index] = {
+                  toolName: "file_search",
+                  toolCallId: value.item.id
+                };
+                controller.enqueue({
+                  type: "tool-input-start",
+                  id: value.item.id,
+                  toolName: "file_search"
+                });
               } else if (value.item.type === "message") {
                 controller.enqueue({
                   type: "text-start",
@@ -3018,6 +3084,32 @@ var OpenAIResponsesLanguageModel = class {
                   },
                   providerExecuted: true
                 });
+              } else if (value.item.type === "file_search_call") {
+                ongoingToolCalls[value.output_index] = void 0;
+                hasToolCalls = true;
+                controller.enqueue({
+                  type: "tool-input-end",
+                  id: value.item.id
+                });
+                controller.enqueue({
+                  type: "tool-call",
+                  toolCallId: value.item.id,
+                  toolName: "file_search",
+                  input: "",
+                  providerExecuted: true
+                });
+                controller.enqueue({
+                  type: "tool-result",
+                  toolCallId: value.item.id,
+                  toolName: "file_search",
+                  result: {
+                    type: "file_search_tool_result",
+                    status: value.item.status || "completed",
+                    ...value.item.queries && { queries: value.item.queries },
+                    ...value.item.results && { results: value.item.results }
+                  },
+                  providerExecuted: true
+                });
               } else if (value.item.type === "message") {
                 controller.enqueue({
                   type: "text-end",
@@ -3196,7 +3288,18 @@ var responseOutputItemAddedSchema = z16.object({
     z16.object({
       type: z16.literal("file_search_call"),
       id: z16.string(),
-      status: z16.string()
+      status: z16.string(),
+      queries: z16.array(z16.string()).nullish(),
+      results: z16.array(
+        z16.object({
+          attributes: z16.object({
+            file_id: z16.string(),
+            filename: z16.string(),
+            score: z16.number(),
+            text: z16.string()
+          })
+        })
+      ).optional()
     })
   ])
 });
@@ -3234,7 +3337,18 @@ var responseOutputItemDoneSchema = z16.object({
     z16.object({
       type: z16.literal("file_search_call"),
       id: z16.string(),
-      status: z16.literal("completed")
+      status: z16.literal("completed"),
+      queries: z16.array(z16.string()).nullish(),
+      results: z16.array(
+        z16.object({
+          attributes: z16.object({
+            file_id: z16.string(),
+            filename: z16.string(),
+            score: z16.number(),
+            text: z16.string()
+          })
+        })
+      ).nullish()
     })
   ])
 });
@@ -3314,6 +3428,13 @@ 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 {
@@ -3335,10 +3456,10 @@ 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 = z16.object({
   metadata: z16.any().nullish(),
@@ -3352,7 +3473,9 @@ var openaiResponsesProviderOptionsSchema = z16.object({
   reasoningSummary: z16.string().nullish(),
   serviceTier: z16.enum(["auto", "flex", "priority"]).nullish(),
   include: z16.array(z16.enum(["reasoning.encrypted_content", "file_search_call.results"])).nullish(),
-  textVerbosity: z16.enum(["low", "medium", "high"]).nullish()
+  textVerbosity: z16.enum(["low", "medium", "high"]).nullish(),
+  promptCacheKey: z16.string().nullish(),
+  safetyIdentifier: z16.string().nullish()
 });
 export {
   OpenAIChatLanguageModel,