feat: support service_tier priority pricing for OpenAI Responses endpoint

Preserve service_tier from request body before field stripping and pass
it through the cost calculation chain. When service_tier is "priority"
and the model has supports_service_tier in pricing data, use *_priority
price fields with automatic fallback to standard pricing.

src/routes/openaiRoutes.js

@@ -283,6 +283,9 @@ const handleResponses = async (req, res) => {
     const codexCliPattern = /^(codex_vscode|codex_cli_rs|codex_exec)\/[\d.]+/i
     const isCodexCLI = codexCliPattern.test(userAgent)
 
+    // 提取 service_tier 用于后续费用计算（在字段被移除前保存）
+    req._serviceTier = req.body?.service_tier || null
+
     // 如果不是 Codex CLI 请求且不是来自 unified 端点（已完成格式转换），则进行适配
     if (!isCodexCLI && !req._fromUnifiedEndpoint) {
       // 移除不需要的请求体字段
@@ -632,7 +635,8 @@ const handleResponses = async (req, res) => {
             cacheReadTokens,
             actualModel,
             accountId,
-            'openai'
+            'openai',
+            req._serviceTier
           )
 
           logger.info(
@@ -749,7 +753,8 @@ const handleResponses = async (req, res) => {
             cacheReadTokens,
             modelToRecord,
             accountId,
-            'openai'
+            'openai',
+            req._serviceTier
           )
 
           logger.info(

src/services/apiKeyService.js

@@ -1539,7 +1539,8 @@ class ApiKeyService {
     cacheReadTokens = 0,
     model = 'unknown',
     accountId = null,
-    accountType = null
+    accountType = null,
+    serviceTier = null
   ) {
     try {
       const totalTokens = inputTokens + outputTokens + cacheCreateTokens + cacheReadTokens
@@ -1553,7 +1554,8 @@ class ApiKeyService {
           cache_creation_input_tokens: cacheCreateTokens,
           cache_read_input_tokens: cacheReadTokens
         },
-        model
+        model,
+        serviceTier
       )
 
       // 检查是否为 1M 上下文请求

src/services/relay/openaiResponsesRelayService.js

@@ -602,6 +602,7 @@ class OpenAIResponsesRelayService {
             usageData.total_tokens || totalInputTokens + outputTokens + cacheCreateTokens
           const modelToRecord = actualModel || requestedModel || 'gpt-4'
 
+          const serviceTier = req._serviceTier || null
           await apiKeyService.recordUsage(
             apiKeyData.id,
             actualInputTokens, // 传递实际输入（不含缓存）
@@ -610,7 +611,8 @@ class OpenAIResponsesRelayService {
             cacheReadTokens,
             modelToRecord,
             account.id,
-            'openai-responses'
+            'openai-responses',
+            serviceTier
           )
 
           logger.info(
@@ -631,7 +633,8 @@ class OpenAIResponsesRelayService {
                 cache_creation_input_tokens: cacheCreateTokens,
                 cache_read_input_tokens: cacheReadTokens
               },
-              modelToRecord
+              modelToRecord,
+              serviceTier
             )
             await openaiResponsesAccountService.updateUsageQuota(account.id, costInfo.costs.total)
           }
@@ -731,6 +734,7 @@ class OpenAIResponsesRelayService {
         const totalTokens =
           usageData.total_tokens || totalInputTokens + outputTokens + cacheCreateTokens
 
+        const serviceTier = req._serviceTier || null
         await apiKeyService.recordUsage(
           apiKeyData.id,
           actualInputTokens, // 传递实际输入（不含缓存）
@@ -739,7 +743,8 @@ class OpenAIResponsesRelayService {
           cacheReadTokens,
           actualModel,
           account.id,
-          'openai-responses'
+          'openai-responses',
+          serviceTier
         )
 
         logger.info(
@@ -760,7 +765,8 @@ class OpenAIResponsesRelayService {
               cache_creation_input_tokens: cacheCreateTokens,
               cache_read_input_tokens: cacheReadTokens
             },
-            actualModel
+            actualModel,
+            serviceTier
           )
           await openaiResponsesAccountService.updateUsageQuota(account.id, costInfo.costs.total)
         }

src/utils/costCalculator.js

@@ -82,7 +82,7 @@ class CostCalculator {
    * @param {string} model - 模型名称
    * @returns {Object} 费用详情
    */
-  static calculateCost(usage, model = 'unknown') {
+  static calculateCost(usage, model = 'unknown', serviceTier = null) {
     // 如果 usage 包含详细的 cache_creation 对象或是 1M 模型，使用 pricingService 来处理
     if (
       (usage.cache_creation && typeof usage.cache_creation === 'object') ||
@@ -148,10 +148,22 @@ class CostCalculator {
     let usingDynamicPricing = false
 
     if (pricingData) {
-      // 转换动态价格格式为内部格式
-      const inputPrice = (pricingData.input_cost_per_token || 0) * 1000000 // 转换为per 1M tokens
-      const outputPrice = (pricingData.output_cost_per_token || 0) * 1000000
-      const cacheReadPrice = (pricingData.cache_read_input_token_cost || 0) * 1000000
+      // 检查是否使用 priority 定价
+      const usePriority = serviceTier === 'priority' && pricingData.supports_service_tier
+
+      // 转换动态价格格式为内部格式（priority 定价时使用 *_priority 字段，回退到标准价格）
+      const inputPrice =
+        ((usePriority && pricingData.input_cost_per_token_priority) ||
+          pricingData.input_cost_per_token ||
+          0) * 1000000
+      const outputPrice =
+        ((usePriority && pricingData.output_cost_per_token_priority) ||
+          pricingData.output_cost_per_token ||
+          0) * 1000000
+      const cacheReadPrice =
+        ((usePriority && pricingData.cache_read_input_token_cost_priority) ||
+          pricingData.cache_read_input_token_cost ||
+          0) * 1000000
 
       // OpenAI 模型的特殊处理：
       // - 如果没有 cache_creation_input_token_cost，缓存创建按普通 input 价格计费