@ai-sdk/openai 4.0.59 → 4.0.60

package/CHANGELOG.md

@@ -1,5 +1,11 @@
 # @ai-sdk/openai
 
+## 4.0.60
+
+### Patch Changes
+
+- 17e489e: feat(openai): add GPT-6 reasoning configuration updates
+
 ## 4.0.59
 
 ### Patch Changes

package/dist/index.d.ts

@@ -10,7 +10,7 @@ declare const openaiLanguageModelChatOptions: _ai_sdk_provider_utils.LazySchema<
     logprobs?: number | boolean | undefined;
     parallelToolCalls?: boolean | undefined;
     user?: string | undefined;
-    reasoningEffort?: "none" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max" | undefined;
+    reasoningEffort?: "low" | "medium" | "high" | "xhigh" | "max" | "none" | "minimal" | undefined;
     maxCompletionTokens?: number | undefined;
     store?: boolean | undefined;
     metadata?: Record<string, string> | undefined;
@@ -49,7 +49,7 @@ type OpenAIEmbeddingModelOptions = InferSchema<typeof openaiEmbeddingModelOption
 
 type OpenAIImageModelId = 'dall-e-3' | 'dall-e-2' | 'gpt-image-1' | 'gpt-image-1-mini' | 'gpt-image-1.5' | 'gpt-image-2' | 'chatgpt-image-latest' | (string & {});
 declare const openaiImageModelOptions: _ai_sdk_provider_utils.LazySchema<{
-    quality?: "auto" | "low" | "medium" | "high" | "standard" | "hd" | undefined;
+    quality?: "low" | "medium" | "high" | "auto" | "standard" | "hd" | undefined;
     background?: "auto" | "transparent" | "opaque" | undefined;
     outputFormat?: "png" | "jpeg" | "webp" | undefined;
     outputCompression?: number | undefined;
@@ -57,17 +57,17 @@ declare const openaiImageModelOptions: _ai_sdk_provider_utils.LazySchema<{
 }>;
 type OpenAIImageModelOptions = InferSchema<typeof openaiImageModelOptions>;
 declare const openaiImageModelGenerationOptions: _ai_sdk_provider_utils.LazySchema<{
-    quality?: "auto" | "low" | "medium" | "high" | "standard" | "hd" | undefined;
+    quality?: "low" | "medium" | "high" | "auto" | "standard" | "hd" | undefined;
     background?: "auto" | "transparent" | "opaque" | undefined;
     outputFormat?: "png" | "jpeg" | "webp" | undefined;
     outputCompression?: number | undefined;
     user?: string | undefined;
     style?: "vivid" | "natural" | undefined;
-    moderation?: "auto" | "low" | undefined;
+    moderation?: "low" | "auto" | undefined;
 }>;
 type OpenAIImageModelGenerationOptions = InferSchema<typeof openaiImageModelGenerationOptions>;
 declare const openaiImageModelEditOptions: _ai_sdk_provider_utils.LazySchema<{
-    quality?: "auto" | "low" | "medium" | "high" | "standard" | "hd" | undefined;
+    quality?: "low" | "medium" | "high" | "auto" | "standard" | "hd" | undefined;
     background?: "auto" | "transparent" | "opaque" | undefined;
     outputFormat?: "png" | "jpeg" | "webp" | undefined;
     outputCompression?: number | undefined;
@@ -1382,6 +1382,7 @@ declare const openaiLanguageModelResponsesOptionsSchema: _ai_sdk_provider_utils.
     } | undefined;
     promptCacheRetention?: "in_memory" | "24h" | null | undefined;
     reasoningEffort?: string | null | undefined;
+    reasoningEffortUpdate?: "low" | "medium" | "high" | "xhigh" | "max" | undefined;
     reasoningMode?: "standard" | "pro" | undefined;
     reasoningContext?: "auto" | "current_turn" | "all_turns" | undefined;
     reasoningSummary?: string | null | undefined;
@@ -1429,7 +1430,7 @@ declare const openAITranscriptionModelOptions: _ai_sdk_provider_utils.LazySchema
         silenceDurationMs?: number | undefined;
     } | undefined;
     streaming?: {
-        delay?: "minimal" | "low" | "medium" | "high" | "xhigh" | undefined;
+        delay?: "low" | "medium" | "high" | "xhigh" | "minimal" | undefined;
         include?: string[] | undefined;
     } | undefined;
 }>;

package/dist/index.js

@@ -48,14 +48,17 @@ function getOpenAILanguageModelCapabilities(modelId) {
   const gptVersion = getGptVersion(modelId);
   const isGptChatModel = (gptVersion == null ? void 0 : gptVersion.minor) == null && ((_b = (_a2 = gptVersion == null ? void 0 : gptVersion.variant) == null ? void 0 : _a2.startsWith("chat")) != null ? _b : false);
   const isGptNanoModel = (_d = (_c = gptVersion == null ? void 0 : gptVersion.variant) == null ? void 0 : _c.startsWith("nano")) != null ? _d : false;
+  const isGpt6OrLaterModel = gptVersion != null && gptVersion.major >= 6;
   const supportsFlexProcessing = oSeriesVersion != null && oSeriesVersion >= 3 || gptVersion != null && gptVersion.major >= 5 && !isGptChatModel;
   const supportsPriorityProcessing = modelId.startsWith("gpt-4") || gptVersion != null && gptVersion.major >= 5 && !isGptNanoModel && !isGptChatModel || oSeriesVersion != null && oSeriesVersion >= 3;
   const isReasoningModel = oSeriesVersion != null || gptVersion != null && gptVersion.major >= 5 && !isGptChatModel;
-  const supportsNonReasoningParameters = gptVersion != null && (gptVersion.major > 5 || gptVersion.major === 5 && ((_e = gptVersion.minor) != null ? _e : 0) >= 1);
+  const supportsNonReasoningParameters = !isGpt6OrLaterModel && gptVersion != null && (gptVersion.major > 5 || gptVersion.major === 5 && ((_e = gptVersion.minor) != null ? _e : 0) >= 1);
   const systemMessageMode = isReasoningModel ? "developer" : "system";
   return {
     supportsFlexProcessing,
     supportsPriorityProcessing,
+    supportsConfigurationUpdate: isGpt6OrLaterModel,
+    supportedReasoningEfforts: isGpt6OrLaterModel ? ["low", "medium", "high", "xhigh", "max"] : void 0,
     isReasoningModel,
     systemMessageMode,
     supportsNonReasoningParameters
@@ -1020,7 +1023,17 @@ var OpenAIChatLanguageModel = class _OpenAIChatLanguageModel {
       schema: openaiLanguageModelChatOptions
     })) != null ? _a2 : {};
     const modelCapabilities = getOpenAILanguageModelCapabilities(this.modelId);
-    const resolvedReasoningEffort = (_b = openaiOptions.reasoningEffort) != null ? _b : isCustomReasoning(reasoning) ? reasoning : void 0;
+    let resolvedReasoningEffort = (_b = openaiOptions.reasoningEffort) != null ? _b : isCustomReasoning(reasoning) ? reasoning : void 0;
+    if (resolvedReasoningEffort != null && modelCapabilities.supportedReasoningEfforts != null && !modelCapabilities.supportedReasoningEfforts.includes(
+      resolvedReasoningEffort
+    )) {
+      warnings.push({
+        type: "unsupported",
+        feature: "reasoningEffort",
+        details: `${this.modelId} only supports the following reasoning efforts: ${modelCapabilities.supportedReasoningEfforts.join(", ")}`
+      });
+      resolvedReasoningEffort = void 0;
+    }
     const isReasoningModel = (_c = openaiOptions.forceReasoning) != null ? _c : modelCapabilities.isReasoningModel;
     if (topK != null) {
       warnings.push({ type: "unsupported", feature: "topK" });
@@ -1075,6 +1088,14 @@ var OpenAIChatLanguageModel = class _OpenAIChatLanguageModel {
       // messages:
       messages
     };
+    if (modelCapabilities.supportedReasoningEfforts != null && baseArgs.prompt_cache_retention != null) {
+      baseArgs.prompt_cache_retention = void 0;
+      warnings.push({
+        type: "unsupported",
+        feature: "promptCacheRetention",
+        details: "promptCacheRetention is not supported by GPT-6 and later models; use promptCacheOptions instead"
+      });
+    }
     if (isReasoningModel) {
       if (resolvedReasoningEffort !== "none" || !modelCapabilities.supportsNonReasoningParameters) {
         if (baseArgs.temperature != null) {
@@ -6073,6 +6094,15 @@ var openaiLanguageModelResponsesOptionsSchema = lazySchema25(
        * Supported values vary by model.
        */
       reasoningEffort: z27.string().nullish(),
+      /**
+       * Updates the reasoning effort for GPT-6 and later models starting with this response
+       * without changing the request-level reasoning effort. This preserves the
+       * request prefix for prompt caching.
+       *
+       * Only supported by GPT-6 and later models in standard, single-agent mode. Cannot be
+       * combined with automatic compaction or automatic truncation.
+       */
+      reasoningEffortUpdate: z27.enum(["low", "medium", "high", "xhigh", "max"]).optional(),
       /**
        * 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.
@@ -6798,7 +6828,7 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
     toolChoice,
     responseFormat
   }) {
-    var _a2, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l;
+    var _a2, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n;
     const warnings = [];
     const modelCapabilities = getOpenAILanguageModelCapabilities(this.modelId);
     if (topK != null) {
@@ -6829,7 +6859,17 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
         schema: openaiLanguageModelResponsesOptionsSchema
       });
     }
-    const resolvedReasoningEffort = (_a2 = openaiOptions == null ? void 0 : openaiOptions.reasoningEffort) != null ? _a2 : isCustomReasoning2(reasoning) ? reasoning : void 0;
+    let resolvedReasoningEffort = (_a2 = openaiOptions == null ? void 0 : openaiOptions.reasoningEffort) != null ? _a2 : isCustomReasoning2(reasoning) ? reasoning : void 0;
+    if (resolvedReasoningEffort != null && modelCapabilities.supportedReasoningEfforts != null && !modelCapabilities.supportedReasoningEfforts.includes(
+      resolvedReasoningEffort
+    )) {
+      warnings.push({
+        type: "unsupported",
+        feature: "reasoningEffort",
+        details: `${this.modelId} only supports the following reasoning efforts: ${modelCapabilities.supportedReasoningEfforts.join(", ")}`
+      });
+      resolvedReasoningEffort = void 0;
+    }
     const resolvedReasoningSummary = (openaiOptions == null ? void 0 : openaiOptions.reasoningSummary) !== void 0 ? openaiOptions.reasoningSummary : resolvedReasoningEffort != null && resolvedReasoningEffort !== "none" ? "detailed" : void 0;
     const isReasoningModel = (_b = openaiOptions == null ? void 0 : openaiOptions.forceReasoning) != null ? _b : modelCapabilities.isReasoningModel;
     if ((openaiOptions == null ? void 0 : openaiOptions.conversation) && (openaiOptions == null ? void 0 : openaiOptions.previousResponseId)) {
@@ -6889,6 +6929,20 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
       outputSchemaToolNames: outputSchemaToolNames.size > 0 ? outputSchemaToolNames : void 0
     });
     warnings.push(...inputWarnings);
+    const reasoningEffortUpdate = openaiOptions == null ? void 0 : openaiOptions.reasoningEffortUpdate;
+    const configurationUpdateIsSupported = reasoningEffortUpdate == null || modelCapabilities.supportsConfigurationUpdate && (openaiOptions == null ? void 0 : openaiOptions.reasoningMode) !== "pro" && (openaiOptions == null ? void 0 : openaiOptions.contextManagement) == null && (openaiOptions == null ? void 0 : openaiOptions.truncation) !== "auto";
+    if (reasoningEffortUpdate != null && !configurationUpdateIsSupported) {
+      warnings.push({
+        type: "unsupported",
+        feature: "reasoningEffortUpdate",
+        details: !modelCapabilities.supportsConfigurationUpdate ? "reasoningEffortUpdate is only supported by GPT-6 and later models" : "reasoningEffortUpdate requires standard reasoning mode without automatic compaction or automatic truncation"
+      });
+    } else if (reasoningEffortUpdate != null) {
+      input.unshift({
+        type: "configuration_update",
+        reasoning: { effort: reasoningEffortUpdate }
+      });
+    }
     if (openaiOptions == null ? void 0 : openaiOptions.compactionTrigger) {
       input.push({ type: "compaction_trigger" });
     }
@@ -6989,6 +7043,14 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
         }
       }
     };
+    if (modelCapabilities.supportsConfigurationUpdate && baseArgs.prompt_cache_retention != null) {
+      baseArgs.prompt_cache_retention = void 0;
+      warnings.push({
+        type: "unsupported",
+        feature: "promptCacheRetention",
+        details: "promptCacheRetention is not supported by GPT-6 and later models; use promptCacheOptions instead"
+      });
+    }
     if (isReasoningModel) {
       if (!(resolvedReasoningEffort === "none" && modelCapabilities.supportsNonReasoningParameters)) {
         if (baseArgs.temperature != null) {
@@ -7007,6 +7069,18 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
             details: "topP is not supported for reasoning models"
           });
         }
+        if (modelCapabilities.supportedReasoningEfforts != null && (baseArgs.top_logprobs != null || ((_j = baseArgs.include) == null ? void 0 : _j.includes("message.output_text.logprobs")))) {
+          baseArgs.top_logprobs = void 0;
+          const filteredInclude = (_k = baseArgs.include) == null ? void 0 : _k.filter(
+            (value) => value !== "message.output_text.logprobs"
+          );
+          baseArgs.include = filteredInclude != null && filteredInclude.length > 0 ? filteredInclude : void 0;
+          warnings.push({
+            type: "unsupported",
+            feature: "logprobs",
+            details: "logprobs is not supported for reasoning models"
+          });
+        }
       }
     } else {
       if ((openaiOptions == null ? void 0 : openaiOptions.reasoningEffort) != null) {
@@ -7054,9 +7128,9 @@ var OpenAIResponsesLanguageModel = class _OpenAIResponsesLanguageModel {
       });
       delete baseArgs.service_tier;
     }
-    const shellToolEnvType = (_l = (_k = (_j = tools == null ? void 0 : tools.find(
+    const shellToolEnvType = (_n = (_m = (_l = tools == null ? void 0 : tools.find(
       (tool) => tool.type === "provider" && tool.id === "openai.shell"
-    )) == null ? void 0 : _j.args) == null ? void 0 : _k.environment) == null ? void 0 : _l.type;
+    )) == null ? void 0 : _l.args) == null ? void 0 : _m.environment) == null ? void 0 : _n.type;
     const isShellProviderExecuted = shellToolEnvType === "containerAuto" || shellToolEnvType === "containerReference";
     return {
       webSearchToolName,
@@ -11060,7 +11134,7 @@ var OpenAISkills = class {
 };
 
 // src/version.ts
-var VERSION = true ? "4.0.59" : "0.0.0-test";
+var VERSION = true ? "4.0.60" : "0.0.0-test";
 
 // src/openai-provider.ts
 function createOpenAI(options = {}) {