✨ feat(gemini): enhance ThinkingAdapter and model handling

- Introduced `isNoThinkingRequest` and `trimModelThinking` functions to manage model names and thinking configurations.
- Updated `GeminiHelper` to conditionally adjust the model name based on the thinking budget and request settings.
- Refactored `ThinkingAdaptor` to streamline the integration of thinking capabilities into Gemini requests.
- Cleaned up commented-out code in `FetchUpstreamModels` for clarity.

These changes improve the handling of model configurations and enhance the adaptability of the Gemini relay system.

controller/channel.go

@@ -134,13 +134,6 @@ func FetchUpstreamModels(c *gin.Context) {
 		return
 	}
 
-	//if channel.Type != common.ChannelTypeOpenAI {
-	//	c.JSON(http.StatusOK, gin.H{
-	//		"success": false,
-	//		"message": "仅支持 OpenAI 类型渠道",
-	//	})
-	//	return
-	//}
 	baseURL := common.ChannelBaseURLs[channel.Type]
 	if channel.GetBaseURL() != "" {
 		baseURL = channel.GetBaseURL()

relay/channel/gemini/relay-gemini.go

@@ -78,26 +78,7 @@ func clampThinkingBudget(modelName string, budget int) int {
 	return budget
 }
 
-// Setting safety to the lowest possible values since Gemini is already powerless enough
+func ThinkingAdaptor(geminiRequest *GeminiChatRequest, info *relaycommon.RelayInfo) {
-func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon.RelayInfo) (*GeminiChatRequest, error) {
-
-	geminiRequest := GeminiChatRequest{
-		Contents: make([]GeminiChatContent, 0, len(textRequest.Messages)),
-		GenerationConfig: GeminiChatGenerationConfig{
-			Temperature:     textRequest.Temperature,
-			TopP:            textRequest.TopP,
-			MaxOutputTokens: textRequest.MaxTokens,
-			Seed:            int64(textRequest.Seed),
-		},
-	}
-
-	if model_setting.IsGeminiModelSupportImagine(info.UpstreamModelName) {
-		geminiRequest.GenerationConfig.ResponseModalities = []string{
-			"TEXT",
-			"IMAGE",
-		}
-	}
-
 	if model_setting.GetGeminiSettings().ThinkingAdapterEnabled {
 		modelName := info.UpstreamModelName
 		isNew25Pro := strings.HasPrefix(modelName, "gemini-2.5-pro") &&
@@ -150,6 +131,29 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 			}
 		}
 	}
+}
+
+// Setting safety to the lowest possible values since Gemini is already powerless enough
+func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon.RelayInfo) (*GeminiChatRequest, error) {
+
+	geminiRequest := GeminiChatRequest{
+		Contents: make([]GeminiChatContent, 0, len(textRequest.Messages)),
+		GenerationConfig: GeminiChatGenerationConfig{
+			Temperature:     textRequest.Temperature,
+			TopP:            textRequest.TopP,
+			MaxOutputTokens: textRequest.MaxTokens,
+			Seed:            int64(textRequest.Seed),
+		},
+	}
+
+	if model_setting.IsGeminiModelSupportImagine(info.UpstreamModelName) {
+		geminiRequest.GenerationConfig.ResponseModalities = []string{
+			"TEXT",
+			"IMAGE",
+		}
+	}
+
+	ThinkingAdaptor(&geminiRequest, info)
 
 	safetySettings := make([]GeminiChatSafetySettings, 0, len(SafetySettingList))
 	for _, category := range SafetySettingList {

relay/gemini_handler.go

@@ -13,6 +13,7 @@ import (
 	"one-api/relay/helper"
 	"one-api/service"
 	"one-api/setting"
+	"one-api/setting/model_setting"
 	"strings"
 
 	"github.com/gin-gonic/gin"
@@ -76,6 +77,33 @@ func getGeminiInputTokens(req *gemini.GeminiChatRequest, info *relaycommon.Relay
 	return inputTokens
 }
 
+func isNoThinkingRequest(req *gemini.GeminiChatRequest) bool {
+	if req.GenerationConfig.ThinkingConfig != nil && req.GenerationConfig.ThinkingConfig.ThinkingBudget != nil {
+		return *req.GenerationConfig.ThinkingConfig.ThinkingBudget <= 0
+	}
+	return false
+}
+
+func trimModelThinking(modelName string) string {
+	// 去除模型名称中的 -nothinking 后缀
+	if strings.HasSuffix(modelName, "-nothinking") {
+		return strings.TrimSuffix(modelName, "-nothinking")
+	}
+	// 去除模型名称中的 -thinking 后缀
+	if strings.HasSuffix(modelName, "-thinking") {
+		return strings.TrimSuffix(modelName, "-thinking")
+	}
+
+	// 去除模型名称中的 -thinking-number
+	if strings.Contains(modelName, "-thinking-") {
+		parts := strings.Split(modelName, "-thinking-")
+		if len(parts) > 1 {
+			return parts[0] + "-thinking"
+		}
+	}
+	return modelName
+}
+
 func GeminiHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
 	req, err := getAndValidateGeminiRequest(c)
 	if err != nil {
@@ -107,12 +135,27 @@ func GeminiHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
 		relayInfo.SetPromptTokens(promptTokens)
 	} else {
 		promptTokens := getGeminiInputTokens(req, relayInfo)
-		if err != nil {
-			return service.OpenAIErrorWrapperLocal(err, "count_input_tokens_error", http.StatusBadRequest)
-		}
 		c.Set("prompt_tokens", promptTokens)
 	}
 
+	if model_setting.GetGeminiSettings().ThinkingAdapterEnabled {
+		if isNoThinkingRequest(req) {
+			// check is thinking
+			if !strings.Contains(relayInfo.OriginModelName, "-nothinking") {
+				// try to get no thinking model price
+				noThinkingModelName := relayInfo.OriginModelName + "-nothinking"
+				containPrice := helper.ContainPriceOrRatio(noThinkingModelName)
+				if containPrice {
+					relayInfo.OriginModelName = noThinkingModelName
+					relayInfo.UpstreamModelName = noThinkingModelName
+				}
+			}
+		}
+		if req.GenerationConfig.ThinkingConfig == nil {
+			gemini.ThinkingAdaptor(req, relayInfo)
+		}
+	}
+
 	priceData, err := helper.ModelPriceHelper(c, relayInfo, relayInfo.PromptTokens, int(req.GenerationConfig.MaxOutputTokens))
 	if err != nil {
 		return service.OpenAIErrorWrapperLocal(err, "model_price_error", http.StatusInternalServerError)