feat: ShouldPreserveThinkingSuffix (#2189)

relay/channel/claude/relay-claude.go

@@ -189,7 +189,9 @@ func RequestOpenAI2ClaudeMessage(c *gin.Context, textRequest dto.GeneralOpenAIRe
 		// https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking#important-considerations-when-using-extended-thinking
 		claudeRequest.TopP = 0
 		claudeRequest.Temperature = common.GetPointer[float64](1.0)
-		claudeRequest.Model = strings.TrimSuffix(textRequest.Model, "-thinking")
+		if !model_setting.ShouldPreserveThinkingSuffix(textRequest.Model) {
+			claudeRequest.Model = strings.TrimSuffix(textRequest.Model, "-thinking")
+		}
 	}
 
 	if textRequest.ReasoningEffort != "" {

relay/channel/gemini/adaptor.go

@@ -127,7 +127,8 @@ func (a *Adaptor) Init(info *relaycommon.RelayInfo) {
 
 func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
 
-	if model_setting.GetGeminiSettings().ThinkingAdapterEnabled {
+	if model_setting.GetGeminiSettings().ThinkingAdapterEnabled &&
+		!model_setting.ShouldPreserveThinkingSuffix(info.OriginModelName) {
 		// 新增逻辑：处理 -thinking-<budget> 格式
 		if strings.Contains(info.UpstreamModelName, "-thinking-") {
 			parts := strings.Split(info.UpstreamModelName, "-thinking-")

relay/channel/openai/adaptor.go

@@ -27,6 +27,7 @@ import (
 	"github.com/QuantumNous/new-api/relay/common_handler"
 	relayconstant "github.com/QuantumNous/new-api/relay/constant"
 	"github.com/QuantumNous/new-api/service"
+	"github.com/QuantumNous/new-api/setting/model_setting"
 	"github.com/QuantumNous/new-api/types"
 
 	"github.com/gin-gonic/gin"
@@ -224,7 +225,8 @@ func (a *Adaptor) ConvertOpenAIRequest(c *gin.Context, info *relaycommon.RelayIn
 			request.Usage = json.RawMessage(`{"include":true}`)
 		}
 		// 适配 OpenRouter 的 thinking 后缀
-		if strings.HasSuffix(info.UpstreamModelName, "-thinking") {
+		if !model_setting.ShouldPreserveThinkingSuffix(info.OriginModelName) &&
+			strings.HasSuffix(info.UpstreamModelName, "-thinking") {
 			info.UpstreamModelName = strings.TrimSuffix(info.UpstreamModelName, "-thinking")
 			request.Model = info.UpstreamModelName
 			if len(request.Reasoning) == 0 {

relay/channel/vertex/adaptor.go

@@ -168,7 +168,8 @@ func (a *Adaptor) getRequestUrl(info *relaycommon.RelayInfo, modelName, suffix s
 func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
 	suffix := ""
 	if a.RequestMode == RequestModeGemini {
-		if model_setting.GetGeminiSettings().ThinkingAdapterEnabled {
+		if model_setting.GetGeminiSettings().ThinkingAdapterEnabled &&
+			!model_setting.ShouldPreserveThinkingSuffix(info.OriginModelName) {
 			// 新增逻辑：处理 -thinking-<budget> 格式
 			if strings.Contains(info.UpstreamModelName, "-thinking-") {
 				parts := strings.Split(info.UpstreamModelName, "-thinking-")

relay/channel/volcengine/adaptor.go

@@ -16,6 +16,7 @@ import (
 	"github.com/QuantumNous/new-api/relay/channel/openai"
 	relaycommon "github.com/QuantumNous/new-api/relay/common"
 	"github.com/QuantumNous/new-api/relay/constant"
+	"github.com/QuantumNous/new-api/setting/model_setting"
 	"github.com/QuantumNous/new-api/types"
 
 	"github.com/gin-gonic/gin"
@@ -291,7 +292,9 @@ func (a *Adaptor) ConvertOpenAIRequest(c *gin.Context, info *relaycommon.RelayIn
 		return nil, errors.New("request is nil")
 	}
 
-	if strings.HasSuffix(info.UpstreamModelName, "-thinking") && strings.HasPrefix(info.UpstreamModelName, "deepseek") {
+	if !model_setting.ShouldPreserveThinkingSuffix(info.OriginModelName) &&
+		strings.HasSuffix(info.UpstreamModelName, "-thinking") &&
+		strings.HasPrefix(info.UpstreamModelName, "deepseek") {
 		info.UpstreamModelName = strings.TrimSuffix(info.UpstreamModelName, "-thinking")
 		request.Model = info.UpstreamModelName
 		request.THINKING = json.RawMessage(`{"type": "enabled"}`)

relay/claude_handler.go

@@ -67,7 +67,9 @@ func ClaudeHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *typ
 			request.TopP = 0
 			request.Temperature = common.GetPointer[float64](1.0)
 		}
-		request.Model = strings.TrimSuffix(request.Model, "-thinking")
+		if !model_setting.ShouldPreserveThinkingSuffix(info.OriginModelName) {
+			request.Model = strings.TrimSuffix(request.Model, "-thinking")
+		}
 		info.UpstreamModelName = request.Model
 	}