@ai-sdk/openai 4.0.10 → 4.0.13

package/dist/internal/index.js

@@ -173,7 +173,7 @@ function isHttpErrorStatusCode(value) {
 
 // src/chat/convert-openai-chat-usage.ts
 function convertOpenAIChatUsage(usage) {
-  var _a, _b, _c, _d, _e, _f;
+  var _a, _b, _c, _d, _e, _f, _g, _h;
   if (usage == null) {
     return {
       inputTokens: {
@@ -193,13 +193,14 @@ function convertOpenAIChatUsage(usage) {
   const promptTokens = (_a = usage.prompt_tokens) != null ? _a : 0;
   const completionTokens = (_b = usage.completion_tokens) != null ? _b : 0;
   const cachedTokens = (_d = (_c = usage.prompt_tokens_details) == null ? void 0 : _c.cached_tokens) != null ? _d : 0;
-  const reasoningTokens = (_f = (_e = usage.completion_tokens_details) == null ? void 0 : _e.reasoning_tokens) != null ? _f : 0;
+  const cacheWriteTokens = (_f = (_e = usage.prompt_tokens_details) == null ? void 0 : _e.cache_write_tokens) != null ? _f : void 0;
+  const reasoningTokens = (_h = (_g = usage.completion_tokens_details) == null ? void 0 : _g.reasoning_tokens) != null ? _h : 0;
   return {
     inputTokens: {
       total: promptTokens,
-      noCache: promptTokens - cachedTokens,
+      noCache: promptTokens - cachedTokens - (cacheWriteTokens != null ? cacheWriteTokens : 0),
       cacheRead: cachedTokens,
-      cacheWrite: void 0
+      cacheWrite: cacheWriteTokens
     },
     outputTokens: {
       total: completionTokens,
@@ -223,23 +224,47 @@ import {
 function serializeToolCallArguments(input) {
   return JSON.stringify(input === void 0 ? {} : input);
 }
+function getPromptCacheBreakpoint(providerOptions) {
+  var _a;
+  return (_a = providerOptions == null ? void 0 : providerOptions.openai) == null ? void 0 : _a.promptCacheBreakpoint;
+}
 function convertToOpenAIChatMessages({
   prompt,
   systemMessageMode = "system"
 }) {
-  var _a;
+  var _a, _b;
   const messages = [];
   const warnings = [];
-  for (const { role, content } of prompt) {
+  for (const { role, content, providerOptions } of prompt) {
     switch (role) {
       case "system": {
         switch (systemMessageMode) {
           case "system": {
-            messages.push({ role: "system", content });
+            const promptCacheBreakpoint = getPromptCacheBreakpoint(providerOptions);
+            messages.push({
+              role: "system",
+              content: promptCacheBreakpoint == null ? content : [
+                {
+                  type: "text",
+                  text: content,
+                  prompt_cache_breakpoint: promptCacheBreakpoint
+                }
+              ]
+            });
             break;
           }
           case "developer": {
-            messages.push({ role: "developer", content });
+            const promptCacheBreakpoint = getPromptCacheBreakpoint(providerOptions);
+            messages.push({
+              role: "developer",
+              content: promptCacheBreakpoint == null ? content : [
+                {
+                  type: "text",
+                  text: content,
+                  prompt_cache_breakpoint: promptCacheBreakpoint
+                }
+              ]
+            });
             break;
           }
           case "remove": {
@@ -259,19 +284,31 @@ function convertToOpenAIChatMessages({
         break;
       }
       case "user": {
-        if (content.length === 1 && content[0].type === "text") {
+        if (content.length === 1 && content[0].type === "text" && getPromptCacheBreakpoint(content[0].providerOptions) == null) {
           messages.push({ role: "user", content: content[0].text });
           break;
         }
         messages.push({
           role: "user",
           content: content.map((part, index) => {
-            var _a2, _b, _c;
+            var _a2, _b2, _c;
             switch (part.type) {
               case "text": {
-                return { type: "text", text: part.text };
+                const promptCacheBreakpoint = getPromptCacheBreakpoint(
+                  part.providerOptions
+                );
+                return {
+                  type: "text",
+                  text: part.text,
+                  ...promptCacheBreakpoint != null && {
+                    prompt_cache_breakpoint: promptCacheBreakpoint
+                  }
+                };
               }
               case "file": {
+                const promptCacheBreakpoint = getPromptCacheBreakpoint(
+                  part.providerOptions
+                );
                 switch (part.data.type) {
                   case "reference": {
                     return {
@@ -281,6 +318,9 @@ function convertToOpenAIChatMessages({
                           reference: part.data.reference,
                           provider: "openai"
                         })
+                      },
+                      ...promptCacheBreakpoint != null && {
+                        prompt_cache_breakpoint: promptCacheBreakpoint
                       }
                     };
                   }
@@ -297,7 +337,10 @@ function convertToOpenAIChatMessages({
                         type: "image_url",
                         image_url: {
                           url: part.data.type === "url" ? part.data.url.toString() : `data:${resolveFullMediaType({ part })};base64,${convertToBase64(part.data.data)}`,
-                          detail: (_b = (_a2 = part.providerOptions) == null ? void 0 : _a2.openai) == null ? void 0 : _b.imageDetail
+                          detail: (_b2 = (_a2 = part.providerOptions) == null ? void 0 : _a2.openai) == null ? void 0 : _b2.imageDetail
+                        },
+                        ...promptCacheBreakpoint != null && {
+                          prompt_cache_breakpoint: promptCacheBreakpoint
                         }
                       };
                     } else if (topLevel === "audio") {
@@ -314,6 +357,9 @@ function convertToOpenAIChatMessages({
                             input_audio: {
                               data: convertToBase64(part.data.data),
                               format: "wav"
+                            },
+                            ...promptCacheBreakpoint != null && {
+                              prompt_cache_breakpoint: promptCacheBreakpoint
                             }
                           };
                         }
@@ -324,6 +370,9 @@ function convertToOpenAIChatMessages({
                             input_audio: {
                               data: convertToBase64(part.data.data),
                               format: "mp3"
+                            },
+                            ...promptCacheBreakpoint != null && {
+                              prompt_cache_breakpoint: promptCacheBreakpoint
                             }
                           };
                         }
@@ -351,6 +400,9 @@ function convertToOpenAIChatMessages({
                         file: {
                           filename: (_c = part.filename) != null ? _c : `part-${index}.pdf`,
                           file_data: `data:application/pdf;base64,${convertToBase64(part.data.data)}`
+                        },
+                        ...promptCacheBreakpoint != null && {
+                          prompt_cache_breakpoint: promptCacheBreakpoint
                         }
                       };
                     }
@@ -364,11 +416,24 @@ function convertToOpenAIChatMessages({
       }
       case "assistant": {
         let text = "";
+        const textParts = [];
+        let hasPromptCacheBreakpoint = false;
         const toolCalls = [];
         for (const part of content) {
           switch (part.type) {
             case "text": {
+              const promptCacheBreakpoint = getPromptCacheBreakpoint(
+                part.providerOptions
+              );
               text += part.text;
+              textParts.push({
+                type: "text",
+                text: part.text,
+                ...promptCacheBreakpoint != null && {
+                  prompt_cache_breakpoint: promptCacheBreakpoint
+                }
+              });
+              hasPromptCacheBreakpoint || (hasPromptCacheBreakpoint = promptCacheBreakpoint != null);
               break;
             }
             case "tool-call": {
@@ -386,7 +451,7 @@ function convertToOpenAIChatMessages({
         }
         messages.push({
           role: "assistant",
-          content: toolCalls.length > 0 ? text || null : text,
+          content: hasPromptCacheBreakpoint ? textParts : toolCalls.length > 0 ? text || null : text,
           tool_calls: toolCalls.length > 0 ? toolCalls : void 0
         });
         break;
@@ -397,6 +462,7 @@ function convertToOpenAIChatMessages({
             continue;
           }
           const output = toolResponse.output;
+          const promptCacheBreakpoint = (_a = output.type === "content" ? output.value.map((part) => getPromptCacheBreakpoint(part.providerOptions)).find((breakpoint) => breakpoint != null) : getPromptCacheBreakpoint(output.providerOptions)) != null ? _a : getPromptCacheBreakpoint(toolResponse.providerOptions);
           let contentValue;
           switch (output.type) {
             case "text":
@@ -404,7 +470,7 @@ function convertToOpenAIChatMessages({
               contentValue = output.value;
               break;
             case "execution-denied":
-              contentValue = (_a = output.reason) != null ? _a : "Tool call execution denied.";
+              contentValue = (_b = output.reason) != null ? _b : "Tool call execution denied.";
               break;
             case "content":
             case "json":
@@ -415,7 +481,13 @@ function convertToOpenAIChatMessages({
           messages.push({
             role: "tool",
             tool_call_id: toolResponse.toolCallId,
-            content: contentValue
+            content: promptCacheBreakpoint == null ? contentValue : [
+              {
+                type: "text",
+                text: contentValue,
+                prompt_cache_breakpoint: promptCacheBreakpoint
+              }
+            ]
           });
         }
         break;
@@ -521,7 +593,8 @@ var openaiChatResponseSchema = lazySchema(
         completion_tokens: z2.number().nullish(),
         total_tokens: z2.number().nullish(),
         prompt_tokens_details: z2.object({
-          cached_tokens: z2.number().nullish()
+          cached_tokens: z2.number().nullish(),
+          cache_write_tokens: z2.number().nullish()
         }).nullish(),
         completion_tokens_details: z2.object({
           reasoning_tokens: z2.number().nullish(),
@@ -590,7 +663,8 @@ var openaiChatChunkSchema = lazySchema(
           completion_tokens: z2.number().nullish(),
           total_tokens: z2.number().nullish(),
           prompt_tokens_details: z2.object({
-            cached_tokens: z2.number().nullish()
+            cached_tokens: z2.number().nullish(),
+            cache_write_tokens: z2.number().nullish()
           }).nullish(),
           completion_tokens_details: z2.object({
             reasoning_tokens: z2.number().nullish(),
@@ -642,7 +716,7 @@ var openaiLanguageModelChatOptions = lazySchema2(
       /**
        * Reasoning effort for reasoning models. Defaults to `medium`.
        */
-      reasoningEffort: z3.enum(["none", "minimal", "low", "medium", "high", "xhigh"]).optional(),
+      reasoningEffort: z3.enum(["none", "minimal", "low", "medium", "high", "xhigh", "max"]).optional(),
       /**
        * Maximum number of completion tokens to generate. Useful for reasoning models.
        */
@@ -686,11 +760,22 @@ var openaiLanguageModelChatOptions = lazySchema2(
        * Useful for improving cache hit rates and working around automatic caching issues.
        */
       promptCacheKey: z3.string().optional(),
+      /**
+       * Prompt cache behavior for GPT-5.6 and later models.
+       * `mode` controls whether OpenAI also places an implicit breakpoint.
+       * `ttl` sets the minimum cache lifetime and currently only supports 30 minutes.
+       */
+      promptCacheOptions: z3.object({
+        mode: z3.enum(["implicit", "explicit"]).optional(),
+        ttl: z3.literal("30m").optional()
+      }).optional(),
       /**
        * The retention policy for the prompt cache.
        * - 'in_memory': Default. Standard prompt caching behavior.
        * - '24h': Extended prompt caching that keeps cached prefixes active for up to 24 hours.
-       *          Currently only available for 5.1 series models.
+       *          Available for models before GPT-5.6 that support extended caching.
+       *
+       * @deprecated For GPT-5.6 and later models, use `promptCacheOptions.ttl`.
        *
        * @default 'in_memory'
        */
@@ -885,6 +970,7 @@ var OpenAIChatLanguageModel = class _OpenAIChatLanguageModel {
       reasoning_effort: resolvedReasoningEffort,
       service_tier: openaiOptions.serviceTier,
       prompt_cache_key: openaiOptions.promptCacheKey,
+      prompt_cache_options: openaiOptions.promptCacheOptions,
       prompt_cache_retention: openaiOptions.promptCacheRetention,
       safety_identifier: openaiOptions.safetyIdentifier,
       // messages:
@@ -2234,16 +2320,16 @@ import {
   convertBase64ToUint8Array as convertBase64ToUint8Array2,
   convertToBase64 as convertToBase642,
   createJsonResponseHandler as createJsonResponseHandler5,
-  getWebSocketConstructor,
+  connectToWebSocket,
   mediaTypeToExtension,
   parseProviderOptions as parseProviderOptions5,
   postFormDataToApi as postFormDataToApi2,
-  readWebSocketMessageText,
   safeParseJSON,
   serializeModelOptions as serializeModelOptions5,
   toWebSocketUrl,
   WORKFLOW_DESERIALIZE as WORKFLOW_DESERIALIZE5,
-  WORKFLOW_SERIALIZE as WORKFLOW_SERIALIZE5
+  WORKFLOW_SERIALIZE as WORKFLOW_SERIALIZE5,
+  waitForWebSocketBufferDrain
 } from "@ai-sdk/provider-utils";
 
 // src/transcription/openai-transcription-api.ts
@@ -2592,21 +2678,18 @@ function createOpenAIRealtimeTranscriptionStream({
   };
   return new ReadableStream({
     start: (controller) => {
-      const WebSocketConstructor = getWebSocketConstructor(webSocket);
-      const ws = new WebSocketConstructor(
-        url,
-        getOpenAIRealtimeProtocols(headers),
-        { headers }
-      );
+      const realtimeConnection = getOpenAIRealtimeConnection(headers);
       let audioReader;
+      let connection;
       cleanup = (closeCode) => {
-        abortSignal == null ? void 0 : abortSignal.removeEventListener("abort", abort);
-        void (audioReader == null ? void 0 : audioReader.cancel().catch(() => {
-        }));
-        try {
-          ws.close(closeCode);
-        } catch (e) {
+        if (audioReader != null) {
+          void audioReader.cancel().catch(() => {
+          });
+        } else {
+          void audio.cancel().catch(() => {
+          });
         }
+        connection == null ? void 0 : connection.close(closeCode);
       };
       const finishWithError = (error) => {
         if (finished) return;
@@ -2629,42 +2712,42 @@ function createOpenAIRealtimeTranscriptionStream({
         controller.close();
         cleanup(1e3);
       };
-      const abort = () => {
-        var _a;
-        finishWithError((_a = abortSignal == null ? void 0 : abortSignal.reason) != null ? _a : new Error("Aborted"));
-      };
-      if (abortSignal == null ? void 0 : abortSignal.aborted) {
-        abort();
-        return;
-      }
-      abortSignal == null ? void 0 : abortSignal.addEventListener("abort", abort, { once: true });
-      const sendAudio = async () => {
+      const sendAudio = async (socket) => {
         audioReader = audio.getReader();
         try {
           while (true) {
             const { done, value } = await audioReader.read();
             if (done || finished) break;
-            ws.send(
+            socket.send(
               JSON.stringify({
                 type: "input_audio_buffer.append",
                 audio: convertToBase642(value)
               })
             );
+            await waitForWebSocketBufferDrain(socket);
           }
         } finally {
           audioReader.releaseLock();
+          audioReader = void 0;
         }
         if (!finished) {
-          ws.send(JSON.stringify({ type: "input_audio_buffer.commit" }));
+          socket.send(JSON.stringify({ type: "input_audio_buffer.commit" }));
         }
       };
-      ws.onopen = () => {
-        controller.enqueue({ type: "stream-start", warnings });
-        ws.send(JSON.stringify(sessionUpdate));
-        void sendAudio().catch(finishWithError);
-      };
-      ws.onmessage = (event) => {
-        void readWebSocketMessageText(event.data).then(async (text) => {
+      connection = connectToWebSocket({
+        url,
+        protocols: realtimeConnection.protocols,
+        headers: realtimeConnection.headers,
+        webSocket,
+        abortSignal,
+        onAbort: finishWithError,
+        onProcessingError: finishWithError,
+        onOpen: (socket) => {
+          controller.enqueue({ type: "stream-start", warnings });
+          socket.send(JSON.stringify(sessionUpdate));
+          void sendAudio(socket).catch(finishWithError);
+        },
+        onMessageText: async (text) => {
           var _a, _b, _c, _d;
           const parsed = await safeParseJSON({ text });
           if (!parsed.success) return;
@@ -2692,17 +2775,17 @@ function createOpenAIRealtimeTranscriptionStream({
               break;
             }
           }
-        }).catch(finishWithError);
-      };
-      ws.onerror = () => {
-        finishWithError(new Error("OpenAI realtime transcription error"));
-      };
-      ws.onclose = () => {
-        if (finished) return;
-        finished = true;
-        cleanup();
-        controller.close();
-      };
+        },
+        onSocketError: () => {
+          finishWithError(new Error("OpenAI realtime transcription error"));
+        },
+        onClose: () => {
+          if (finished) return;
+          finished = true;
+          cleanup();
+          controller.close();
+        }
+      });
     },
     cancel: () => {
       if (finished) return;
@@ -2739,11 +2822,26 @@ function buildOpenAIRealtimeTranscriptionSession({
     }
   };
 }
-function getOpenAIRealtimeProtocols(headers) {
+function getOpenAIRealtimeConnection(headers) {
   var _a;
-  const authorization = (_a = headers.Authorization) != null ? _a : headers.authorization;
-  const token = (authorization == null ? void 0 : authorization.startsWith("Bearer ")) ? authorization.slice("Bearer ".length) : void 0;
-  return token == null ? ["realtime"] : ["realtime", `openai-insecure-api-key.${token}`];
+  let authorization;
+  for (const [key, value] of Object.entries(headers)) {
+    if (key.toLowerCase() === "authorization" && value != null) {
+      authorization = value;
+    }
+  }
+  const token = (_a = authorization == null ? void 0 : authorization.match(/^bearer\s+(.+)$/i)) == null ? void 0 : _a[1];
+  if (token == null) {
+    return { protocols: ["realtime"], headers };
+  }
+  return {
+    protocols: ["realtime", `openai-insecure-api-key.${token}`],
+    headers: Object.fromEntries(
+      Object.entries(headers).filter(
+        ([key]) => key.toLowerCase() !== "authorization"
+      )
+    )
+  };
 }
 
 // src/speech/openai-speech-model.ts
@@ -2902,7 +3000,7 @@ import {
 
 // src/responses/convert-openai-responses-usage.ts
 function convertOpenAIResponsesUsage(usage) {
-  var _a, _b, _c, _d;
+  var _a, _b, _c, _d, _e, _f;
   if (usage == null) {
     return {
       inputTokens: {
@@ -2922,13 +3020,14 @@ function convertOpenAIResponsesUsage(usage) {
   const inputTokens = usage.input_tokens;
   const outputTokens = usage.output_tokens;
   const cachedTokens = (_b = (_a = usage.input_tokens_details) == null ? void 0 : _a.cached_tokens) != null ? _b : 0;
-  const reasoningTokens = (_d = (_c = usage.output_tokens_details) == null ? void 0 : _c.reasoning_tokens) != null ? _d : 0;
+  const cacheWriteTokens = (_d = (_c = usage.input_tokens_details) == null ? void 0 : _c.cache_write_tokens) != null ? _d : void 0;
+  const reasoningTokens = (_f = (_e = usage.output_tokens_details) == null ? void 0 : _e.reasoning_tokens) != null ? _f : 0;
   return {
     inputTokens: {
       total: inputTokens,
-      noCache: inputTokens - cachedTokens,
+      noCache: inputTokens - cachedTokens - (cacheWriteTokens != null ? cacheWriteTokens : 0),
       cacheRead: cachedTokens,
-      cacheWrite: void 0
+      cacheWrite: cacheWriteTokens
     },
     outputTokens: {
       total: outputTokens,
@@ -3173,6 +3272,10 @@ var toolSearchToolFactory = createProviderDefinedToolFactoryWithOutputSchema4({
 function serializeToolCallArguments2(input) {
   return JSON.stringify(input === void 0 ? {} : input);
 }
+function getPromptCacheBreakpoint2(providerOptions, providerOptionsName) {
+  var _a;
+  return (_a = providerOptions == null ? void 0 : providerOptions[providerOptionsName]) == null ? void 0 : _a.promptCacheBreakpoint;
+}
 function isFileId(data, prefixes) {
   if (!prefixes) return false;
   return prefixes.some((prefix) => data.startsWith(prefix));
@@ -3196,16 +3299,42 @@ async function convertToOpenAIResponsesInput({
   let input = [];
   const warnings = [];
   const processedApprovalIds = /* @__PURE__ */ new Set();
-  for (const { role, content } of prompt) {
+  for (const { role, content, providerOptions } of prompt) {
     switch (role) {
       case "system": {
         switch (systemMessageMode) {
           case "system": {
-            input.push({ role: "system", content });
+            const promptCacheBreakpoint = getPromptCacheBreakpoint2(
+              providerOptions,
+              providerOptionsName
+            );
+            input.push({
+              role: "system",
+              content: promptCacheBreakpoint == null ? content : [
+                {
+                  type: "input_text",
+                  text: content,
+                  prompt_cache_breakpoint: promptCacheBreakpoint
+                }
+              ]
+            });
             break;
           }
           case "developer": {
-            input.push({ role: "developer", content });
+            const promptCacheBreakpoint = getPromptCacheBreakpoint2(
+              providerOptions,
+              providerOptionsName
+            );
+            input.push({
+              role: "developer",
+              content: promptCacheBreakpoint == null ? content : [
+                {
+                  type: "input_text",
+                  text: content,
+                  prompt_cache_breakpoint: promptCacheBreakpoint
+                }
+              ]
+            });
             break;
           }
           case "remove": {
@@ -3231,9 +3360,23 @@ async function convertToOpenAIResponsesInput({
             var _a2, _b2, _c2, _d2, _e2;
             switch (part.type) {
               case "text": {
-                return { type: "input_text", text: part.text };
+                const promptCacheBreakpoint = getPromptCacheBreakpoint2(
+                  part.providerOptions,
+                  providerOptionsName
+                );
+                return {
+                  type: "input_text",
+                  text: part.text,
+                  ...promptCacheBreakpoint != null && {
+                    prompt_cache_breakpoint: promptCacheBreakpoint
+                  }
+                };
               }
               case "file": {
+                const promptCacheBreakpoint = getPromptCacheBreakpoint2(
+                  part.providerOptions,
+                  providerOptionsName
+                );
                 switch (part.data.type) {
                   case "reference": {
                     const fileId = resolveProviderReference2({
@@ -3244,12 +3387,18 @@ async function convertToOpenAIResponsesInput({
                       return {
                         type: "input_image",
                         file_id: fileId,
-                        detail: (_b2 = (_a2 = part.providerOptions) == null ? void 0 : _a2[providerOptionsName]) == null ? void 0 : _b2.imageDetail
+                        detail: (_b2 = (_a2 = part.providerOptions) == null ? void 0 : _a2[providerOptionsName]) == null ? void 0 : _b2.imageDetail,
+                        ...promptCacheBreakpoint != null && {
+                          prompt_cache_breakpoint: promptCacheBreakpoint
+                        }
                       };
                     }
                     return {
                       type: "input_file",
-                      file_id: fileId
+                      file_id: fileId,
+                      ...promptCacheBreakpoint != null && {
+                        prompt_cache_breakpoint: promptCacheBreakpoint
+                      }
                     };
                   }
                   case "text": {
@@ -3266,13 +3415,19 @@ async function convertToOpenAIResponsesInput({
                         ...part.data.type === "url" ? { image_url: part.data.url.toString() } : typeof part.data.data === "string" && isFileId(part.data.data, fileIdPrefixes) ? { file_id: part.data.data } : {
                           image_url: `data:${resolveFullMediaType2({ part })};base64,${convertToBase643(part.data.data)}`
                         },
-                        detail: (_d2 = (_c2 = part.providerOptions) == null ? void 0 : _c2[providerOptionsName]) == null ? void 0 : _d2.imageDetail
+                        detail: (_d2 = (_c2 = part.providerOptions) == null ? void 0 : _c2[providerOptionsName]) == null ? void 0 : _d2.imageDetail,
+                        ...promptCacheBreakpoint != null && {
+                          prompt_cache_breakpoint: promptCacheBreakpoint
+                        }
                       };
                     } else {
                       if (part.data.type === "url") {
                         return {
                           type: "input_file",
-                          file_url: part.data.url.toString()
+                          file_url: part.data.url.toString(),
+                          ...promptCacheBreakpoint != null && {
+                            prompt_cache_breakpoint: promptCacheBreakpoint
+                          }
                         };
                       }
                       const fullMediaType = resolveFullMediaType2({ part });
@@ -3286,6 +3441,9 @@ async function convertToOpenAIResponsesInput({
                         ...typeof part.data.data === "string" && isFileId(part.data.data, fileIdPrefixes) ? { file_id: part.data.data } : {
                           filename: (_e2 = part.filename) != null ? _e2 : fullMediaType === "application/pdf" ? `part-${index}.pdf` : `part-${index}`,
                           file_data: `data:${fullMediaType};base64,${convertToBase643(part.data.data)}`
+                        },
+                        ...promptCacheBreakpoint != null && {
+                          prompt_cache_breakpoint: promptCacheBreakpoint
                         }
                       };
                     }
@@ -3302,9 +3460,9 @@ async function convertToOpenAIResponsesInput({
         for (const part of content) {
           switch (part.type) {
             case "text": {
-              const providerOptions = (_a = part.providerOptions) == null ? void 0 : _a[providerOptionsName];
-              const id = providerOptions == null ? void 0 : providerOptions.itemId;
-              const phase = providerOptions == null ? void 0 : providerOptions.phase;
+              const providerOptions2 = (_a = part.providerOptions) == null ? void 0 : _a[providerOptionsName];
+              const id = providerOptions2 == null ? void 0 : providerOptions2.itemId;
+              const phase = providerOptions2 == null ? void 0 : providerOptions2.phase;
               if (hasConversation && id != null) {
                 break;
               }
@@ -3501,12 +3659,12 @@ async function convertToOpenAIResponsesInput({
               break;
             }
             case "reasoning": {
-              const providerOptions = await parseProviderOptions7({
+              const providerOptions2 = await parseProviderOptions7({
                 provider: providerOptionsName,
                 providerOptions: part.providerOptions,
                 schema: openaiResponsesReasoningProviderOptionsSchema
               });
-              const reasoningId = providerOptions == null ? void 0 : providerOptions.itemId;
+              const reasoningId = providerOptions2 == null ? void 0 : providerOptions2.itemId;
               if ((hasConversation || hasPreviousResponseId) && reasoningId != null) {
                 break;
               }
@@ -3538,19 +3696,19 @@ async function convertToOpenAIResponsesInput({
                     reasoningMessages[reasoningId] = {
                       type: "reasoning",
                       id: reasoningId,
-                      encrypted_content: providerOptions == null ? void 0 : providerOptions.reasoningEncryptedContent,
+                      encrypted_content: providerOptions2 == null ? void 0 : providerOptions2.reasoningEncryptedContent,
                       summary: summaryParts
                     };
                     input.push(reasoningMessages[reasoningId]);
                   } else {
                     reasoningMessage.summary.push(...summaryParts);
-                    if ((providerOptions == null ? void 0 : providerOptions.reasoningEncryptedContent) != null) {
-                      reasoningMessage.encrypted_content = providerOptions.reasoningEncryptedContent;
+                    if ((providerOptions2 == null ? void 0 : providerOptions2.reasoningEncryptedContent) != null) {
+                      reasoningMessage.encrypted_content = providerOptions2.reasoningEncryptedContent;
                     }
                   }
                 }
               } else {
-                const encryptedContent = providerOptions == null ? void 0 : providerOptions.reasoningEncryptedContent;
+                const encryptedContent = providerOptions2 == null ? void 0 : providerOptions2.reasoningEncryptedContent;
                 if (encryptedContent != null) {
                   const summaryParts = [];
                   if (part.text.length > 0) {
@@ -3575,8 +3733,8 @@ async function convertToOpenAIResponsesInput({
             }
             case "custom": {
               if (part.kind === "openai.compaction") {
-                const providerOptions = (_t = part.providerOptions) == null ? void 0 : _t[providerOptionsName];
-                const id = providerOptions == null ? void 0 : providerOptions.itemId;
+                const providerOptions2 = (_t = part.providerOptions) == null ? void 0 : _t[providerOptionsName];
+                const id = providerOptions2 == null ? void 0 : providerOptions2.itemId;
                 if (hasConversation && id != null) {
                   break;
                 }
@@ -3584,7 +3742,7 @@ async function convertToOpenAIResponsesInput({
                   input.push({ type: "item_reference", id });
                   break;
                 }
-                const encryptedContent = providerOptions == null ? void 0 : providerOptions.encryptedContent;
+                const encryptedContent = providerOptions2 == null ? void 0 : providerOptions2.encryptedContent;
                 if (id != null) {
                   input.push({
                     type: "compaction",
@@ -3705,9 +3863,19 @@ async function convertToOpenAIResponsesInput({
               case "content":
                 outputValue = output.value.map((item) => {
                   var _a2, _b2, _c2;
+                  const promptCacheBreakpoint = getPromptCacheBreakpoint2(
+                    item.providerOptions,
+                    providerOptionsName
+                  );
                   switch (item.type) {
                     case "text":
-                      return { type: "input_text", text: item.text };
+                      return {
+                        type: "input_text",
+                        text: item.text,
+                        ...promptCacheBreakpoint != null && {
+                          prompt_cache_breakpoint: promptCacheBreakpoint
+                        }
+                      };
                     case "file": {
                       const topLevel = getTopLevelMediaType2(item.mediaType);
                       const imageDetail = (_b2 = (_a2 = item.providerOptions) == null ? void 0 : _a2[providerOptionsName]) == null ? void 0 : _b2.imageDetail;
@@ -3719,13 +3887,19 @@ async function convertToOpenAIResponsesInput({
                           return {
                             type: "input_image",
                             image_url: `data:${fullMediaType};base64,${convertToBase643(item.data.data)}`,
-                            detail: imageDetail
+                            detail: imageDetail,
+                            ...promptCacheBreakpoint != null && {
+                              prompt_cache_breakpoint: promptCacheBreakpoint
+                            }
                           };
                         }
                         return {
                           type: "input_file",
                           filename: (_c2 = item.filename) != null ? _c2 : "data",
-                          file_data: `data:${fullMediaType};base64,${convertToBase643(item.data.data)}`
+                          file_data: `data:${fullMediaType};base64,${convertToBase643(item.data.data)}`,
+                          ...promptCacheBreakpoint != null && {
+                            prompt_cache_breakpoint: promptCacheBreakpoint
+                          }
                         };
                       }
                       if (item.data.type === "url") {
@@ -3733,12 +3907,18 @@ async function convertToOpenAIResponsesInput({
                           return {
                             type: "input_image",
                             image_url: item.data.url.toString(),
-                            detail: imageDetail
+                            detail: imageDetail,
+                            ...promptCacheBreakpoint != null && {
+                              prompt_cache_breakpoint: promptCacheBreakpoint
+                            }
                           };
                         }
                         return {
                           type: "input_file",
-                          file_url: item.data.url.toString()
+                          file_url: item.data.url.toString(),
+                          ...promptCacheBreakpoint != null && {
+                            prompt_cache_breakpoint: promptCacheBreakpoint
+                          }
                         };
                       }
                       warnings.push({
@@ -3782,9 +3962,19 @@ async function convertToOpenAIResponsesInput({
             case "content":
               contentValue = output.value.map((item) => {
                 var _a2, _b2, _c2;
+                const promptCacheBreakpoint = getPromptCacheBreakpoint2(
+                  item.providerOptions,
+                  providerOptionsName
+                );
                 switch (item.type) {
                   case "text": {
-                    return { type: "input_text", text: item.text };
+                    return {
+                      type: "input_text",
+                      text: item.text,
+                      ...promptCacheBreakpoint != null && {
+                        prompt_cache_breakpoint: promptCacheBreakpoint
+                      }
+                    };
                   }
                   case "file": {
                     const topLevel = getTopLevelMediaType2(item.mediaType);
@@ -3797,13 +3987,19 @@ async function convertToOpenAIResponsesInput({
                         return {
                           type: "input_image",
                           image_url: `data:${fullMediaType};base64,${convertToBase643(item.data.data)}`,
-                          detail: imageDetail
+                          detail: imageDetail,
+                          ...promptCacheBreakpoint != null && {
+                            prompt_cache_breakpoint: promptCacheBreakpoint
+                          }
                         };
                       }
                       return {
                         type: "input_file",
                         filename: (_c2 = item.filename) != null ? _c2 : "data",
-                        file_data: `data:${fullMediaType};base64,${convertToBase643(item.data.data)}`
+                        file_data: `data:${fullMediaType};base64,${convertToBase643(item.data.data)}`,
+                        ...promptCacheBreakpoint != null && {
+                          prompt_cache_breakpoint: promptCacheBreakpoint
+                        }
                       };
                     }
                     if (item.data.type === "url") {
@@ -3811,12 +4007,18 @@ async function convertToOpenAIResponsesInput({
                         return {
                           type: "input_image",
                           image_url: item.data.url.toString(),
-                          detail: imageDetail
+                          detail: imageDetail,
+                          ...promptCacheBreakpoint != null && {
+                            prompt_cache_breakpoint: promptCacheBreakpoint
+                          }
                         };
                       }
                       return {
                         type: "input_file",
-                        file_url: item.data.url.toString()
+                        file_url: item.data.url.toString(),
+                        ...promptCacheBreakpoint != null && {
+                          prompt_cache_breakpoint: promptCacheBreakpoint
+                        }
                       };
                     }
                     warnings.push({
@@ -3947,6 +4149,7 @@ var openaiResponsesChunkSchema = lazySchema16(
             input_tokens: z18.number(),
             input_tokens_details: z18.object({
               cached_tokens: z18.number().nullish(),
+              cache_write_tokens: z18.number().nullish(),
               orchestration_input_tokens: z18.number().nullish(),
               orchestration_input_cached_tokens: z18.number().nullish()
             }).nullish(),
@@ -3956,6 +4159,9 @@ var openaiResponsesChunkSchema = lazySchema16(
               orchestration_output_tokens: z18.number().nullish()
             }).nullish()
           }),
+          reasoning: z18.object({
+            context: z18.string().nullish()
+          }).nullish(),
           service_tier: z18.string().nullish()
         })
       }),
@@ -3972,6 +4178,7 @@ var openaiResponsesChunkSchema = lazySchema16(
             input_tokens: z18.number(),
             input_tokens_details: z18.object({
               cached_tokens: z18.number().nullish(),
+              cache_write_tokens: z18.number().nullish(),
               orchestration_input_tokens: z18.number().nullish(),
               orchestration_input_cached_tokens: z18.number().nullish()
             }).nullish(),
@@ -3981,6 +4188,9 @@ var openaiResponsesChunkSchema = lazySchema16(
               orchestration_output_tokens: z18.number().nullish()
             }).nullish()
           }).nullish(),
+          reasoning: z18.object({
+            context: z18.string().nullish()
+          }).nullish(),
           service_tier: z18.string().nullish()
         })
       }),
@@ -4767,11 +4977,15 @@ var openaiResponsesResponseSchema = lazySchema16(
         ])
       ).optional(),
       service_tier: z18.string().nullish(),
+      reasoning: z18.object({
+        context: z18.string().nullish()
+      }).nullish(),
       incomplete_details: z18.object({ reason: z18.string() }).nullish(),
       usage: z18.object({
         input_tokens: z18.number(),
         input_tokens_details: z18.object({
           cached_tokens: z18.number().nullish(),
+          cache_write_tokens: z18.number().nullish(),
           orchestration_input_tokens: z18.number().nullish(),
           orchestration_input_cached_tokens: z18.number().nullish()
         }).nullish(),
@@ -4929,11 +5143,22 @@ var openaiLanguageModelResponsesOptionsSchema = lazySchema17(
        * Sets a cache key to tie this prompt to cached prefixes for better caching performance.
        */
       promptCacheKey: z19.string().nullish(),
+      /**
+       * Prompt cache behavior for GPT-5.6 and later models.
+       * `mode` controls whether OpenAI also places an implicit breakpoint.
+       * `ttl` sets the minimum cache lifetime and currently only supports 30 minutes.
+       */
+      promptCacheOptions: z19.object({
+        mode: z19.enum(["implicit", "explicit"]).optional(),
+        ttl: z19.literal("30m").optional()
+      }).optional(),
       /**
        * The retention policy for the prompt cache.
        * - 'in_memory': Default. Standard prompt caching behavior.
        * - '24h': Extended prompt caching that keeps cached prefixes active for up to 24 hours.
-       *          Currently only available for 5.1 series models.
+       *          Available for models before GPT-5.6 that support extended caching.
+       *
+       * @deprecated For GPT-5.6 and later models, use `promptCacheOptions.ttl`.
        *
        * @default 'in_memory'
        */
@@ -4941,14 +5166,21 @@ var openaiLanguageModelResponsesOptionsSchema = lazySchema17(
       /**
        * Reasoning effort for reasoning models. Defaults to `medium`. If you use
        * `providerOptions` to set the `reasoningEffort` option, this model setting will be ignored.
-       * Valid values: 'none' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh'
-       *
-       * The 'none' type for `reasoningEffort` is only available for OpenAI's GPT-5.1
-       * models. Also, the 'xhigh' type for `reasoningEffort` is only available for
-       * OpenAI's GPT-5.1-Codex-Max model. Setting `reasoningEffort` to 'none' or 'xhigh' with unsupported models will result in
-       * an error.
+       * GPT-5.6 supports 'none' | 'low' | 'medium' | 'high' | 'xhigh' | 'max'.
+       * Supported values vary by model.
        */
       reasoningEffort: z19.string().nullish(),
+      /**
+       * Controls how much model work GPT-5.6 performs before returning a final answer.
+       * `standard` is the default. `pro` increases quality, latency, and token usage.
+       */
+      reasoningMode: z19.enum(["standard", "pro"]).optional(),
+      /**
+       * Controls which available reasoning items GPT-5.6 can use.
+       * `auto` uses the model default, `current_turn` excludes reasoning from earlier
+       * turns, and `all_turns` makes compatible earlier reasoning available.
+       */
+      reasoningContext: z19.enum(["auto", "current_turn", "all_turns"]).optional(),
       /**
        * Controls reasoning summary output from the model.
        * Set to "auto" to automatically receive the richest level available,
@@ -5929,6 +6161,7 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
       service_tier: openaiOptions == null ? void 0 : openaiOptions.serviceTier,
       include,
       prompt_cache_key: openaiOptions == null ? void 0 : openaiOptions.promptCacheKey,
+      prompt_cache_options: openaiOptions == null ? void 0 : openaiOptions.promptCacheOptions,
       prompt_cache_retention: openaiOptions == null ? void 0 : openaiOptions.promptCacheRetention,
       safety_identifier: openaiOptions == null ? void 0 : openaiOptions.safetyIdentifier,
       top_logprobs: topLogprobs,
@@ -5941,13 +6174,19 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
         }))
       },
       // model-specific settings:
-      ...isReasoningModel && (resolvedReasoningEffort != null || resolvedReasoningSummary != null) && {
+      ...isReasoningModel && (resolvedReasoningEffort != null || resolvedReasoningSummary != null || (openaiOptions == null ? void 0 : openaiOptions.reasoningMode) != null || (openaiOptions == null ? void 0 : openaiOptions.reasoningContext) != null) && {
         reasoning: {
           ...resolvedReasoningEffort != null && {
             effort: resolvedReasoningEffort
           },
           ...resolvedReasoningSummary != null && {
             summary: resolvedReasoningSummary
+          },
+          ...(openaiOptions == null ? void 0 : openaiOptions.reasoningMode) != null && {
+            mode: openaiOptions.reasoningMode
+          },
+          ...(openaiOptions == null ? void 0 : openaiOptions.reasoningContext) != null && {
+            context: openaiOptions.reasoningContext
           }
         }
       }
@@ -5986,6 +6225,20 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
           details: "reasoningSummary is not supported for non-reasoning models"
         });
       }
+      if ((openaiOptions == null ? void 0 : openaiOptions.reasoningMode) != null) {
+        warnings.push({
+          type: "unsupported",
+          feature: "reasoningMode",
+          details: "reasoningMode is not supported for non-reasoning models"
+        });
+      }
+      if ((openaiOptions == null ? void 0 : openaiOptions.reasoningContext) != null) {
+        warnings.push({
+          type: "unsupported",
+          feature: "reasoningContext",
+          details: "reasoningContext is not supported for non-reasoning models"
+        });
+      }
     }
     if ((openaiOptions == null ? void 0 : openaiOptions.serviceTier) === "flex" && !modelCapabilities.supportsFlexProcessing) {
       warnings.push({
@@ -6022,7 +6275,7 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
     };
   }
   async doGenerate(options) {
-    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, _y, _z, _A, _B, _C, _D;
+    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, _y, _z, _A, _B, _C, _D, _E;
     const {
       args: body,
       warnings,
@@ -6485,7 +6738,8 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
       [providerOptionsName]: {
         responseId: response.id,
         ...logprobs.length > 0 ? { logprobs } : {},
-        ...typeof response.service_tier === "string" ? { serviceTier: response.service_tier } : {}
+        ...typeof response.service_tier === "string" ? { serviceTier: response.service_tier } : {},
+        ...((_B = response.reasoning) == null ? void 0 : _B.context) != null ? { reasoningContext: response.reasoning.context } : {}
       }
     };
     const usage = response.usage;
@@ -6493,10 +6747,10 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
       content,
       finishReason: {
         unified: mapOpenAIResponseFinishReason({
-          finishReason: (_B = response.incomplete_details) == null ? void 0 : _B.reason,
+          finishReason: (_C = response.incomplete_details) == null ? void 0 : _C.reason,
           hasFunctionCall
         }),
-        raw: (_D = (_C = response.incomplete_details) == null ? void 0 : _C.reason) != null ? _D : void 0
+        raw: (_E = (_D = response.incomplete_details) == null ? void 0 : _D.reason) != null ? _E : void 0
       },
       usage: convertOpenAIResponsesUsage(usage),
       request: { body },
@@ -6564,6 +6818,7 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
     let hasFunctionCall = false;
     const activeReasoning = {};
     let serviceTier;
+    let reasoningContext;
     const hostedToolSearchCallIds = [];
     let encounteredStreamError = false;
     const result = {
@@ -6573,7 +6828,7 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
             controller.enqueue({ type: "stream-start", warnings });
           },
           transform(chunk, controller) {
-            var _a2, _b2, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w, _x, _y, _z, _A, _B, _C, _D, _E, _F, _G, _H, _I, _J, _K, _L;
+            var _a2, _b2, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w, _x, _y, _z, _A, _B, _C, _D, _E, _F, _G, _H, _I, _J, _K, _L, _M, _N;
             if (options.includeRawChunks) {
               controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
             }
@@ -7327,8 +7582,11 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
               if (typeof value.response.service_tier === "string") {
                 serviceTier = value.response.service_tier;
               }
+              if (((_y = value.response.reasoning) == null ? void 0 : _y.context) != null) {
+                reasoningContext = value.response.reasoning.context;
+              }
             } else if (isResponseFailedChunk(value)) {
-              const incompleteReason = (_y = value.response.incomplete_details) == null ? void 0 : _y.reason;
+              const incompleteReason = (_z = value.response.incomplete_details) == null ? void 0 : _z.reason;
               finishReason = {
                 unified: incompleteReason ? mapOpenAIResponseFinishReason({
                   finishReason: incompleteReason,
@@ -7336,7 +7594,10 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
                 }) : "error",
                 raw: incompleteReason != null ? incompleteReason : "error"
               };
-              usage = (_z = value.response.usage) != null ? _z : void 0;
+              usage = (_A = value.response.usage) != null ? _A : void 0;
+              if (((_B = value.response.reasoning) == null ? void 0 : _B.context) != null) {
+                reasoningContext = value.response.reasoning.context;
+              }
               if (!encounteredStreamError && value.response.error != null) {
                 encounteredStreamError = true;
                 controller.enqueue({
@@ -7358,7 +7619,7 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
                 controller.enqueue({
                   type: "source",
                   sourceType: "url",
-                  id: (_C = (_B = (_A = self.config).generateId) == null ? void 0 : _B.call(_A)) != null ? _C : generateId2(),
+                  id: (_E = (_D = (_C = self.config).generateId) == null ? void 0 : _D.call(_C)) != null ? _E : generateId2(),
                   url: value.annotation.url,
                   title: value.annotation.title
                 });
@@ -7366,7 +7627,7 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
                 controller.enqueue({
                   type: "source",
                   sourceType: "document",
-                  id: (_F = (_E = (_D = self.config).generateId) == null ? void 0 : _E.call(_D)) != null ? _F : generateId2(),
+                  id: (_H = (_G = (_F = self.config).generateId) == null ? void 0 : _G.call(_F)) != null ? _H : generateId2(),
                   mediaType: "text/plain",
                   title: value.annotation.filename,
                   filename: value.annotation.filename,
@@ -7382,7 +7643,7 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
                 controller.enqueue({
                   type: "source",
                   sourceType: "document",
-                  id: (_I = (_H = (_G = self.config).generateId) == null ? void 0 : _H.call(_G)) != null ? _I : generateId2(),
+                  id: (_K = (_J = (_I = self.config).generateId) == null ? void 0 : _J.call(_I)) != null ? _K : generateId2(),
                   mediaType: "text/plain",
                   title: value.annotation.filename,
                   filename: value.annotation.filename,
@@ -7398,7 +7659,7 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
                 controller.enqueue({
                   type: "source",
                   sourceType: "document",
-                  id: (_L = (_K = (_J = self.config).generateId) == null ? void 0 : _K.call(_J)) != null ? _L : generateId2(),
+                  id: (_N = (_M = (_L = self.config).generateId) == null ? void 0 : _M.call(_L)) != null ? _N : generateId2(),
                   mediaType: "application/octet-stream",
                   title: value.annotation.file_id,
                   filename: value.annotation.file_id,
@@ -7422,7 +7683,8 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
               [providerOptionsName]: {
                 responseId,
                 ...logprobs.length > 0 ? { logprobs } : {},
-                ...serviceTier !== void 0 ? { serviceTier } : {}
+                ...serviceTier !== void 0 ? { serviceTier } : {},
+                ...reasoningContext !== void 0 ? { reasoningContext } : {}
               }
             };
             controller.enqueue({