@openclaw/ai 2026.7.2-beta.1 → 2026.7.2-beta.3

package/dist/{mistral-D-39Trax.mjs → mistral-BdrQmDWp.mjs}

@@ -1,10 +1,10 @@
 import { n as getEnvApiKey } from "./env-api-keys-DnhYpc27.mjs";
 import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
 import { n as getAiTransportHost } from "./host-4t713IeR.mjs";
-import { n as calculateCost, r as clampThinkingLevel } from "./model-utils-Q1LSRIdo.mjs";
+import { n as calculateCost, r as clampThinkingLevel } from "./model-utils-1GiZ2_rr.mjs";
 import { n as parseStreamingJson } from "./json-parse-BvXNt1-7.mjs";
 import { t as sanitizeSurrogates } from "./sanitize-unicode-DT5o51ur.mjs";
-import { a as truncateUtf16Safe, i as extractToolResultText, n as describeToolResultMediaPlaceholder, s as buildBaseOptions, t as transformMessages, x as stripSystemPromptCacheBoundary } from "./transform-messages-DNWpwvXA.mjs";
+import { i as extractToolResultText, l as buildBaseOptions, n as describeToolResultMediaPlaceholder, o as isImageWithMediaPayload, s as truncateUtf16Safe, t as transformMessages, u as clampMaxTokensToModel, w as stripSystemPromptCacheBoundary } from "./transform-messages-eb6lxDLB.mjs";
 import { t as shortHash } from "./hash-CHgqbJmD.mjs";
 import { t as createSseByteGuard } from "./streaming-byte-guard-BrbkbwUu.mjs";
 import { randomUUID } from "node:crypto";
@@ -75,7 +75,7 @@ const streamMistral = (model, context, options) => {
 				...model.headers,
 				...options?.headers
 			};
-			if (options?.sessionId) headers["x-affinity"] ||= options.sessionId;
+			if (resolveMistralPromptCacheKey(options) && options?.sessionId) headers["x-affinity"] ||= options.sessionId;
 			const mistralStream = await mistral.chat.stream(payload, {
 				headers,
 				signal: options?.signal
@@ -113,7 +113,10 @@ const streamMistral = (model, context, options) => {
 const streamSimpleMistral = (model, context, options) => {
 	const apiKey = options?.apiKey || getEnvApiKey(model.provider);
 	if (!apiKey) throw new Error(`No API key for provider: ${model.provider}`);
-	const base = buildBaseOptions(model, options, apiKey);
+	const base = {
+		...buildBaseOptions(model, options, apiKey),
+		maxTokens: clampMaxTokensToModel(model, options?.maxTokens)
+	};
 	const clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : void 0;
 	const reasoning = clampedReasoning === "off" ? void 0 : clampedReasoning;
 	const shouldUseReasoning = model.reasoning && reasoning !== void 0;
@@ -218,12 +221,23 @@ function buildChatPayload(model, context, messages, options) {
 	}
 	if (options?.promptMode) payload.promptMode = options.promptMode;
 	if (options?.reasoningEffort) payload.reasoningEffort = options.reasoningEffort;
+	const promptCacheKey = resolveMistralPromptCacheKey(options);
+	if (promptCacheKey) payload.promptCacheKey = promptCacheKey;
 	if (context.systemPrompt) payload.messages.unshift({
 		role: "system",
 		content: sanitizeSurrogates(stripSystemPromptCacheBoundary(context.systemPrompt))
 	});
 	return payload;
 }
+function resolveMistralPromptCacheKey(options) {
+	if (options?.cacheRetention === "none") return;
+	return options?.promptCacheKey?.trim() || options?.sessionId?.trim() || void 0;
+}
+function readMistralCachedPromptTokens(usage, promptTokens) {
+	const record = usage;
+	const rawCachedTokens = record.promptTokensDetails?.cachedTokens ?? record.prompt_tokens_details?.cached_tokens ?? record.cachedTokens ?? record.cached_tokens;
+	return Math.min(promptTokens, Math.max(0, typeof rawCachedTokens === "number" && Number.isFinite(rawCachedTokens) ? rawCachedTokens : 0));
+}
 async function consumeChatStream(model, output, stream, mistralStream) {
 	let currentBlock = null;
 	const blocks = output.content;
@@ -309,11 +323,13 @@ async function consumeChatStream(model, output, stream, mistralStream) {
 		const chunk = event.data;
 		output.responseId ||= chunk.id;
 		if (chunk.usage) {
-			output.usage.input = chunk.usage.promptTokens || 0;
+			const promptTokens = chunk.usage.promptTokens || 0;
+			const cachedPromptTokens = readMistralCachedPromptTokens(chunk.usage, promptTokens);
+			output.usage.input = Math.max(0, promptTokens - cachedPromptTokens);
 			output.usage.output = chunk.usage.completionTokens || 0;
-			output.usage.cacheRead = 0;
+			output.usage.cacheRead = cachedPromptTokens;
 			output.usage.cacheWrite = 0;
-			output.usage.totalTokens = chunk.usage.totalTokens || output.usage.input + output.usage.output;
+			output.usage.totalTokens = chunk.usage.totalTokens || output.usage.input + output.usage.output + output.usage.cacheRead;
 			calculateCost(model, output.usage);
 		}
 		const choice = chunk.choices[0];
@@ -576,14 +592,14 @@ function toChatMessages(messages, supportsImages) {
 			continue;
 		}
 		const toolContent = [];
-		const toolText = buildToolResultText(extractToolResultText(msg.content), describeToolResultMediaPlaceholder(msg.content), msg.content.some((part) => part.type === "image"), supportsImages, msg.isError);
+		const toolText = buildToolResultText(extractToolResultText(msg.content), describeToolResultMediaPlaceholder(msg.content), msg.content.some(isImageWithMediaPayload), supportsImages, msg.isError);
 		toolContent.push({
 			type: "text",
 			text: toolText
 		});
 		for (const part of msg.content) {
 			if (!supportsImages) continue;
-			if (part.type !== "image") continue;
+			if (!isImageWithMediaPayload(part)) continue;
 			toolContent.push({
 				type: "image_url",
 				imageUrl: `data:${part.mimeType};base64,${part.data}`

package/dist/{model-utils-Q1LSRIdo.mjs → model-utils-1GiZ2_rr.mjs}

@@ -2,10 +2,12 @@ import { c as resolveClaudeNativeThinkingLevelMap, i as requiresClaudeMandatoryA
 //#region packages/ai/src/model-utils.ts
 /** Calculates and stores model cost fields from token usage and per-million pricing. */
 function calculateCost(model, usage) {
+	const cacheWrite1h = Math.min(usage.cacheWrite, Math.max(0, usage.cacheWrite1h ?? 0));
+	const cacheWrite5m = usage.cacheWrite - cacheWrite1h;
 	usage.cost.input = model.cost.input / 1e6 * usage.input;
 	usage.cost.output = model.cost.output / 1e6 * usage.output;
 	usage.cost.cacheRead = model.cost.cacheRead / 1e6 * usage.cacheRead;
-	usage.cost.cacheWrite = model.cost.cacheWrite / 1e6 * usage.cacheWrite;
+	usage.cost.cacheWrite = (model.cost.cacheWrite * cacheWrite5m + model.cost.input * 2 * cacheWrite1h) / 1e6;
 	usage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;
 	return usage.cost;
 }

package/dist/{openai-chatgpt-responses-DCA0K85L.mjs → openai-chatgpt-responses-DP63Qt01.mjs}

@@ -3,10 +3,10 @@ import { i as formatThrownValue, n as createAssistantMessageDiagnostic, t as app
 import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
 import { n as getAiTransportHost, r as resolveAiTransportHeaderSentinels } from "./host-4t713IeR.mjs";
 import { t as headersToRecord } from "./headers-B_e4-1J0.mjs";
-import { s as buildBaseOptions, x as stripSystemPromptCacheBoundary } from "./transform-messages-DNWpwvXA.mjs";
-import { A as supportsOpenAITemperature, a as resolveResponsesReasoningEffort, i as processResponsesStream, n as convertResponsesMessages, s as convertResponsesToolPayload } from "./openai-responses-shared-B9mtGvXq.mjs";
+import { l as buildBaseOptions, w as stripSystemPromptCacheBoundary } from "./transform-messages-eb6lxDLB.mjs";
+import { M as supportsOpenAITemperature, a as resolveResponsesReasoningEffort, i as processResponsesStream, n as convertResponsesMessages, s as convertResponsesToolPayload } from "./openai-responses-shared-CwnGU0KW.mjs";
 import { i as getFirstStreamEventTimeoutMs, o as clampTimerTimeoutMs, r as getFirstStreamEventTimeoutHandler, s as resolveTimerTimeoutMs, t as createFirstStreamEventAbortController } from "./stream-first-event-timeout-RjWszj8c.mjs";
-import { a as clampOpenAIPromptCacheKey } from "./openai-tool-projection-cTLC4F65.mjs";
+import { a as clampOpenAIPromptCacheKey } from "./openai-tool-projection-WHf2nuOI.mjs";
 import { parseRetryAfterHttpDateMs } from "./internal/retry-after.mjs";
 import { i as registerSessionResourceCleanup, n as resolveOpenAICodexAccountId } from "./openai-chatgpt-jwt-DhAAzLkj.mjs";
 import { t as createSseByteGuard } from "./streaming-byte-guard-BrbkbwUu.mjs";
@@ -38,7 +38,7 @@ function loadNodeOs() {
 }
 const os = loadNodeOs();
 const DEFAULT_CODEX_BASE_URL = "https://chatgpt.com/backend-api";
-const MAX_RETRIES = 3;
+const DEFAULT_MAX_RETRIES = 3;
 const BASE_DELAY_MS = 1e3;
 const REQUEST_COMPRESSION_ZSTD_LEVEL = 3;
 const CODEX_TOOL_CALL_PROVIDERS = /* @__PURE__ */ new Set(["openai", "opencode"]);
@@ -127,8 +127,9 @@ const streamOpenAICodexResponses = (model, context, options) => {
 			let body = buildRequestBody(model, context, options);
 			const nextBody = await options?.onPayload?.(body, model);
 			if (nextBody !== void 0) body = nextBody;
-			const websocketRequestId = options?.sessionId || createCodexRequestId();
-			const sseHeaders = buildSSEHeaders(modelHeaders, optionHeaders, accountId, apiKey, options?.sessionId);
+			const sessionId = clampOpenAIPromptCacheKey(options?.sessionId);
+			const websocketRequestId = sessionId || createCodexRequestId();
+			const sseHeaders = buildSSEHeaders(modelHeaders, optionHeaders, accountId, apiKey, sessionId);
 			const websocketHeaders = buildWebSocketHeaders(modelHeaders, optionHeaders, accountId, apiKey, websocketRequestId);
 			const bodyJson = JSON.stringify(body);
 			requestTimeoutMs = resolveRequestTimeoutMs(options);
@@ -184,7 +185,8 @@ const streamOpenAICodexResponses = (model, context, options) => {
 			const sseBody = compressedBody ?? bodyJson;
 			let response;
 			let lastError;
-			for (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {
+			const maxRetries = options?.maxRetries ?? DEFAULT_MAX_RETRIES;
+			for (let attempt = 0; attempt <= maxRetries; attempt++) {
 				if (activeSignal?.aborted) throw new Error("Request was aborted");
 				try {
 					response = await fetch(resolveCodexUrl(model.baseUrl), {
@@ -199,7 +201,7 @@ const streamOpenAICodexResponses = (model, context, options) => {
 					}, model);
 					if (response.ok) break;
 					const errorText = await readChatGptResponsesErrorTextLimited(response);
-					if (attempt < MAX_RETRIES && isRetryableError(response.status, errorText)) {
+					if (attempt < maxRetries && isRetryableError(response.status, errorText)) {
 						let delayMs = BASE_DELAY_MS * 2 ** attempt;
 						const retryAfterMs = response.headers.get("retry-after-ms");
 						if (retryAfterMs !== null) {
@@ -233,7 +235,7 @@ const streamOpenAICodexResponses = (model, context, options) => {
 						if (error.name === "TimeoutError" && requestTimeoutMs !== void 0) throw new Error(`Request timed out after ${requestTimeoutMs}ms`, { cause: error });
 					}
 					lastError = error instanceof Error ? error : new Error(String(error));
-					if (attempt < MAX_RETRIES && !lastError.message.includes("usage limit")) {
+					if (attempt < maxRetries && !lastError.message.includes("usage limit")) {
 						await sleepWithAbort(BASE_DELAY_MS * 2 ** attempt, activeSignal);
 						continue;
 					}

package/dist/{openai-completions-B2jr3vtJ.mjs → openai-completions-zLDfebnN.mjs}

@@ -1,15 +1,16 @@
 import { n as getEnvApiKey, r as __exportAll } from "./env-api-keys-DnhYpc27.mjs";
 import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
 import { n as getAiTransportHost } from "./host-4t713IeR.mjs";
-import { n as calculateCost, r as clampThinkingLevel, t as applyProviderReportedUsageCost } from "./model-utils-Q1LSRIdo.mjs";
+import { n as calculateCost, r as clampThinkingLevel, t as applyProviderReportedUsageCost } from "./model-utils-1GiZ2_rr.mjs";
 import { t as headersToRecord } from "./headers-B_e4-1J0.mjs";
 import { n as parseStreamingJson } from "./json-parse-BvXNt1-7.mjs";
 import { t as sanitizeSurrogates } from "./sanitize-unicode-DT5o51ur.mjs";
-import { b as splitSystemPromptCacheBoundary, i as extractToolResultText, n as describeToolResultMediaPlaceholder, s as buildBaseOptions, t as transformMessages, x as stripSystemPromptCacheBoundary } from "./transform-messages-DNWpwvXA.mjs";
+import { C as splitSystemPromptCacheBoundary, i as extractToolResultText, l as buildBaseOptions, n as describeToolResultMediaPlaceholder, o as isImageWithMediaPayload, t as transformMessages, w as stripSystemPromptCacheBoundary } from "./transform-messages-eb6lxDLB.mjs";
 import { a as resolveCacheRetention, i as resolveCloudflareBaseUrl, n as hasCopilotVisionInput, r as isCloudflareProvider, t as buildCopilotDynamicHeaders } from "./github-copilot-headers-BsH5cqGj.mjs";
+import { t as formatProviderError } from "./provider-error-LMfTEkfO.mjs";
 import { t as createReasoningTagTextPartitioner } from "./reasoning-tag-text-partitioner-BlVe67g6.mjs";
 import { a as withFirstStreamEventTimeout, i as getFirstStreamEventTimeoutMs, r as getFirstStreamEventTimeoutHandler, t as createFirstStreamEventAbortController } from "./stream-first-event-timeout-RjWszj8c.mjs";
-import { a as clampOpenAIPromptCacheKey, n as reconcileOpenAICompletionsToolChoice, t as projectOpenAITools } from "./openai-tool-projection-cTLC4F65.mjs";
+import { a as clampOpenAIPromptCacheKey, n as reconcileOpenAICompletionsToolChoice, t as projectOpenAITools } from "./openai-tool-projection-WHf2nuOI.mjs";
 import OpenAI from "openai";
 //#region packages/ai/src/providers/openai-stop-reason.ts
 function mapOpenAIStopReason(reason, options) {
@@ -67,14 +68,16 @@ function isThinkingContentBlock(block) {
 function isToolCallBlock(block) {
 	return block.type === "toolCall";
 }
-function isImageContentBlock(block) {
-	return block.type === "image";
-}
 const EMPTY_TOOL_RESULT_TEXT = "(no output)";
 function sanitizeToolResultText(text, fallback) {
 	const sanitized = sanitizeSurrogates(text);
 	return sanitized.trim().length > 0 ? sanitized : fallback;
 }
+function isEncryptedReasoningDetail(detail) {
+	if (typeof detail !== "object" || detail === null) return false;
+	const candidate = detail;
+	return candidate.type === "reasoning.encrypted" && typeof candidate.id === "string" && candidate.id.length > 0 && typeof candidate.data === "string" && candidate.data.length > 0;
+}
 const streamOpenAICompletions = (model, context, options) => {
 	const stream = new AssistantMessageEventStream();
 	(async () => {
@@ -115,7 +118,7 @@ const streamOpenAICompletions = (model, context, options) => {
 			const requestOptions = {
 				signal: firstEventAbort.signal,
 				...options?.timeoutMs !== void 0 ? { timeout: options.timeoutMs } : {},
-				...options?.maxRetries !== void 0 ? { maxRetries: options.maxRetries } : {}
+				maxRetries: options?.maxRetries ?? 0
 			};
 			const { data: openaiStream, response } = await client.chat.completions.create(params, requestOptions).withResponse();
 			await options?.onResponse?.({
@@ -132,6 +135,7 @@ const streamOpenAICompletions = (model, context, options) => {
 			const toolCallBlocksByIndex = /* @__PURE__ */ new Map();
 			const toolCallBlocksById = /* @__PURE__ */ new Map();
 			const toolCallBlocksByFirstId = /* @__PURE__ */ new Map();
+			const pendingReasoningDetailsByToolCallId = /* @__PURE__ */ new Map();
 			const blocks = output.content;
 			const finishedBlocks = /* @__PURE__ */ new Set();
 			const contentIndices = /* @__PURE__ */ new WeakMap();
@@ -141,7 +145,13 @@ const streamOpenAICompletions = (model, context, options) => {
 			};
 			const getContentIndex = (block) => contentIndices.get(block) ?? -1;
 			const rememberFirstToolCallById = (id, block) => {
-				if (!toolCallBlocksByFirstId.has(id)) toolCallBlocksByFirstId.set(id, block);
+				if (toolCallBlocksByFirstId.has(id)) return;
+				toolCallBlocksByFirstId.set(id, block);
+				const pendingDetail = pendingReasoningDetailsByToolCallId.get(id);
+				if (pendingDetail) {
+					block.thoughtSignature = pendingDetail;
+					pendingReasoningDetailsByToolCallId.delete(id);
+				}
 			};
 			const finishBlock = (block) => {
 				const contentIndex = getContentIndex(block);
@@ -344,10 +354,12 @@ const streamOpenAICompletions = (model, context, options) => {
 						}
 					}
 					const reasoningDetails = choiceDelta.reasoning_details;
-					if (reasoningDetails && Array.isArray(reasoningDetails)) {
-						for (const detail of reasoningDetails) if (detail.type === "reasoning.encrypted" && detail.id && detail.data) {
+					if (Array.isArray(reasoningDetails)) {
+						for (const detail of reasoningDetails) if (isEncryptedReasoningDetail(detail)) {
+							const serializedDetail = JSON.stringify(detail);
 							const matchingToolCall = toolCallBlocksByFirstId.get(detail.id);
-							if (matchingToolCall) matchingToolCall.thoughtSignature = JSON.stringify(detail);
+							if (matchingToolCall) matchingToolCall.thoughtSignature = serializedDetail;
+							else pendingReasoningDetailsByToolCallId.set(detail.id, serializedDetail);
 						}
 					}
 				}
@@ -376,9 +388,9 @@ const streamOpenAICompletions = (model, context, options) => {
 				delete block.streamIndex;
 			}
 			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
-			output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
+			output.errorMessage = formatProviderError(error);
 			const rawMetadata = error?.error?.metadata?.raw;
-			if (rawMetadata) output.errorMessage += `\n${rawMetadata}`;
+			if (rawMetadata && !output.errorMessage.includes(rawMetadata)) output.errorMessage += `\n${rawMetadata}`;
 			stream.push({
 				type: "error",
 				reason: output.stopReason,
@@ -415,7 +427,8 @@ function createClient(model, context, apiKey, optionsHeaders, sessionId, compat
 		});
 		Object.assign(headers, copilotHeaders);
 	}
-	if (sessionId && compat.sendSessionAffinityHeaders) {
+	if (sessionId && compat.sendSessionAffinityHeaders) if (compat.sessionAffinityFormat === "openrouter") headers["x-session-id"] = sessionId;
+	else {
 		headers.session_id = sessionId;
 		headers["x-client-request-id"] = sessionId;
 		headers["x-session-affinity"] = sessionId;
@@ -472,7 +485,10 @@ function buildParams(model, context, options, compat = getCompat(model), cacheRe
 		const toolChoice = reconcileOpenAICompletionsToolChoice(options.toolChoice, toolProjection ?? projectOpenAITools([]));
 		if (toolChoice !== void 0) params.tool_choice = toolChoice;
 	}
-	if (compat.thinkingFormat === "zai" && model.reasoning) params.enable_thinking = Boolean(options?.reasoningEffort);
+	if (compat.thinkingFormat === "zai" && model.reasoning) params.thinking = options?.reasoningEffort ? {
+		type: "enabled",
+		clear_thinking: false
+	} : { type: "disabled" };
 	else if (compat.thinkingFormat === "qwen" && model.reasoning) params.enable_thinking = Boolean(options?.reasoningEffort);
 	else if (compat.thinkingFormat === "qwen-chat-template" && model.reasoning) params.chat_template_kwargs = {
 		enable_thinking: Boolean(options?.reasoningEffort),
@@ -494,7 +510,7 @@ function buildParams(model, context, options, compat = getCompat(model), cacheRe
 		const offValue = model.thinkingLevelMap?.off;
 		if (typeof offValue === "string") params.reasoning_effort = offValue;
 	}
-	if (model.baseUrl.includes("openrouter.ai") && model.compat?.openRouterRouting) params.provider = model.compat.openRouterRouting;
+	if (model.compat?.openRouterRouting) params.provider = model.compat.openRouterRouting;
 	if (model.baseUrl.includes("ai-gateway.vercel.sh") && model.compat?.vercelGatewayRouting) {
 		const routing = model.compat.vercelGatewayRouting;
 		if (routing.only || routing.order) {
@@ -699,7 +715,7 @@ function convertMessages(model, context, compat, options = {}) {
 				if (toolMsg?.role !== "toolResult") break;
 				const textResult = extractToolResultText(toolMsg.content);
 				const mediaPlaceholder = describeToolResultMediaPlaceholder(toolMsg.content);
-				const hasImages = toolMsg.content.some((c) => c.type === "image");
+				const hasImages = toolMsg.content.some(isImageWithMediaPayload);
 				const toolResultMsg = {
 					role: "tool",
 					content: sanitizeToolResultText(textResult, mediaPlaceholder ?? EMPTY_TOOL_RESULT_TEXT),
@@ -708,7 +724,7 @@ function convertMessages(model, context, compat, options = {}) {
 				if (compat.requiresToolResultName && toolMsg.toolName) toolResultMsg.name = toolMsg.toolName;
 				params.push(toolResultMsg);
 				if (hasImages && model.input.includes("image")) {
-					for (const block of toolMsg.content) if (isImageContentBlock(block)) imageBlocks.push({
+					for (const block of toolMsg.content) if (isImageWithMediaPayload(block)) imageBlocks.push({
 						type: "image_url",
 						image_url: { url: `data:${block.mimeType};base64,${block.data}` }
 					});
@@ -788,15 +804,18 @@ function detectCompat(model) {
 	const isMoonshot = provider === "moonshotai" || provider === "moonshotai-cn" || baseUrl.includes("api.moonshot.");
 	const isCloudflareWorkersAI = provider === "cloudflare-workers-ai" || baseUrl.includes("api.cloudflare.com");
 	const isCloudflareAiGateway = provider === "cloudflare-ai-gateway" || baseUrl.includes("gateway.ai.cloudflare.com");
+	const isOpenRouter = provider === "openrouter" || baseUrl.includes("openrouter.ai");
 	const isNonStandard = provider === "cerebras" || baseUrl.includes("cerebras.ai") || provider === "xai" || baseUrl.includes("api.x.ai") || isTogether || baseUrl.includes("chutes.ai") || baseUrl.includes("deepseek.com") || isZai || isMoonshot || provider === "opencode" || baseUrl.includes("opencode.ai") || isCloudflareWorkersAI || isCloudflareAiGateway;
-	const useMaxTokens = baseUrl.includes("chutes.ai") || isMoonshot || isCloudflareAiGateway || isTogether;
+	const useMaxTokens = baseUrl.includes("chutes.ai") || isMoonshot || isCloudflareAiGateway || isTogether || isZai;
 	const isGrok = provider === "xai" || baseUrl.includes("api.x.ai");
 	const isDeepSeek = provider === "deepseek" || baseUrl.includes("deepseek.com");
 	const isXiaomi = provider === "xiaomi" || baseUrl.includes("xiaomimimo.com");
+	const supportsOpenRouterDeveloperRole = isOpenRouter && (model.id.startsWith("anthropic/") || model.id.startsWith("openai/"));
+	const usesOpenRouterSessionAffinity = isOpenRouter || model.compat?.thinkingFormat === "openrouter" || model.compat?.openRouterRouting !== void 0;
 	const cacheControlFormat = provider === "openrouter" && model.id.startsWith("anthropic/") ? "anthropic" : void 0;
 	return {
 		supportsStore: !isNonStandard,
-		supportsDeveloperRole: !isNonStandard,
+		supportsDeveloperRole: supportsOpenRouterDeveloperRole || !isNonStandard && !isOpenRouter,
 		supportsReasoningEffort: !isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway,
 		supportsUsageInStreaming: true,
 		maxTokensField: useMaxTokens ? "max_tokens" : "max_completion_tokens",
@@ -804,13 +823,14 @@ function detectCompat(model) {
 		requiresAssistantAfterToolResult: false,
 		requiresThinkingAsText: false,
 		requiresReasoningContentOnAssistantMessages: isDeepSeek || isXiaomi,
-		thinkingFormat: isDeepSeek ? "deepseek" : isXiaomi ? "deepseek" : isZai ? "zai" : isTogether ? "together" : provider === "openrouter" || baseUrl.includes("openrouter.ai") ? "openrouter" : "openai",
+		thinkingFormat: isDeepSeek ? "deepseek" : isXiaomi ? "deepseek" : isZai ? "zai" : isTogether ? "together" : isOpenRouter ? "openrouter" : "openai",
 		openRouterRouting: {},
 		vercelGatewayRouting: {},
 		zaiToolStream: false,
 		supportsStrictMode: !isMoonshot && !isTogether && !isCloudflareAiGateway,
 		cacheControlFormat,
 		sendSessionAffinityHeaders: false,
+		sessionAffinityFormat: usesOpenRouterSessionAffinity ? "openrouter" : "openai",
 		supportsPromptCacheKey: false,
 		supportsLongCacheRetention: !(isTogether || isCloudflareWorkersAI || isCloudflareAiGateway)
 	};
@@ -839,6 +859,7 @@ function getCompat(model) {
 		supportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,
 		cacheControlFormat: model.compat.cacheControlFormat ?? detected.cacheControlFormat,
 		sendSessionAffinityHeaders: model.compat.sendSessionAffinityHeaders ?? detected.sendSessionAffinityHeaders,
+		sessionAffinityFormat: detected.sessionAffinityFormat,
 		supportsPromptCacheKey: model.compat.supportsPromptCacheKey ?? detected.supportsPromptCacheKey,
 		supportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? detected.supportsLongCacheRetention
 	};

package/dist/{openai-responses-DTpD8YVy.mjs → openai-responses-ClspWt4I.mjs}

@@ -1,10 +1,10 @@
 import { n as getEnvApiKey, r as __exportAll } from "./env-api-keys-DnhYpc27.mjs";
 import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
 import { n as getAiTransportHost } from "./host-4t713IeR.mjs";
-import { s as buildBaseOptions } from "./transform-messages-DNWpwvXA.mjs";
+import { l as buildBaseOptions } from "./transform-messages-eb6lxDLB.mjs";
 import { a as resolveCacheRetention, i as resolveCloudflareBaseUrl, n as hasCopilotVisionInput, r as isCloudflareProvider, t as buildCopilotDynamicHeaders } from "./github-copilot-headers-BsH5cqGj.mjs";
-import { A as supportsOpenAITemperature, a as resolveResponsesReasoningEffort, n as convertResponsesMessages, o as runResponsesStreamLifecycle, r as createResponsesAssistantOutput, t as applyCommonResponsesParams } from "./openai-responses-shared-B9mtGvXq.mjs";
-import { a as clampOpenAIPromptCacheKey } from "./openai-tool-projection-cTLC4F65.mjs";
+import { M as supportsOpenAITemperature, a as resolveResponsesReasoningEffort, n as convertResponsesMessages, o as runResponsesStreamLifecycle, r as createResponsesAssistantOutput, t as applyCommonResponsesParams } from "./openai-responses-shared-CwnGU0KW.mjs";
+import { a as clampOpenAIPromptCacheKey } from "./openai-tool-projection-WHf2nuOI.mjs";
 import OpenAI from "openai";
 //#region packages/ai/src/providers/openai-responses.ts
 var openai_responses_exports = /* @__PURE__ */ __exportAll({