Merge pull request #1247 from RedwindA/feat/25lite-thinking

feat: improve gemini thinking budget adaption

common/redis.go

@@ -141,7 +141,11 @@ func RedisHSetObj(key string, obj interface{}, expiration time.Duration) error {
 
 	txn := RDB.TxPipeline()
 	txn.HSet(ctx, key, data)
-	txn.Expire(ctx, key, expiration)
+
+	// 只有在 expiration 大于 0 时才设置过期时间
+	if expiration > 0 {
+		txn.Expire(ctx, key, expiration)
+	}
 
 	_, err := txn.Exec(ctx)
 	if err != nil {

common/utils.go

@@ -249,13 +249,38 @@ func SaveTmpFile(filename string, data io.Reader) (string, error) {
 }
 
 // GetAudioDuration returns the duration of an audio file in seconds.
-func GetAudioDuration(ctx context.Context, filename string) (float64, error) {
+func GetAudioDuration(ctx context.Context, filename string, ext string) (float64, error) {
 	// ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 {{input}}
 	c := exec.CommandContext(ctx, "ffprobe", "-v", "error", "-show_entries", "format=duration", "-of", "default=noprint_wrappers=1:nokey=1", filename)
 	output, err := c.Output()
 	if err != nil {
 		return 0, errors.Wrap(err, "failed to get audio duration")
 	}
+  durationStr := string(bytes.TrimSpace(output))
+  if durationStr == "N/A" {
+    // Create a temporary output file name
+    tmpFp, err := os.CreateTemp("", "audio-*"+ext)
+    if err != nil {
+      return 0, errors.Wrap(err, "failed to create temporary file")
+    }
+    tmpName := tmpFp.Name()
+    // Close immediately so ffmpeg can open the file on Windows.
+    _ = tmpFp.Close()
+    defer os.Remove(tmpName)
 
-	return strconv.ParseFloat(string(bytes.TrimSpace(output)), 64)
+    // ffmpeg -y -i filename -vcodec copy -acodec copy <tmpName>
+    ffmpegCmd := exec.CommandContext(ctx, "ffmpeg", "-y", "-i", filename, "-vcodec", "copy", "-acodec", "copy", tmpName)
+    if err := ffmpegCmd.Run(); err != nil {
+      return 0, errors.Wrap(err, "failed to run ffmpeg")
+    }
+
+    // Recalculate the duration of the new file
+    c = exec.CommandContext(ctx, "ffprobe", "-v", "error", "-show_entries", "format=duration", "-of", "default=noprint_wrappers=1:nokey=1", tmpName)
+    output, err := c.Output()
+    if err != nil {
+      return 0, errors.Wrap(err, "failed to get audio duration after ffmpeg")
+    }
+    durationStr = string(bytes.TrimSpace(output))
+  }
+	return strconv.ParseFloat(durationStr, 64)
 }

constant/cache_key.go

@@ -2,12 +2,10 @@ package constant
 
 import "one-api/common"
 
-var (
-	TokenCacheSeconds         = common.SyncFrequency
-	UserId2GroupCacheSeconds  = common.SyncFrequency
-	UserId2QuotaCacheSeconds  = common.SyncFrequency
-	UserId2StatusCacheSeconds = common.SyncFrequency
-)
+// 使用函数来避免初始化顺序带来的赋值问题
+func RedisKeyCacheSeconds() int {
+	return common.SyncFrequency
+}
 
 // Cache keys
 const (

controller/channel-test.go

@@ -165,8 +165,8 @@ func testChannel(channel *model.Channel, testModel string) (err error, openAIErr
 	tok := time.Now()
 	milliseconds := tok.Sub(tik).Milliseconds()
 	consumedTime := float64(milliseconds) / 1000.0
-	other := service.GenerateTextOtherInfo(c, info, priceData.ModelRatio, priceData.GroupRatio, priceData.CompletionRatio,
-		usage.PromptTokensDetails.CachedTokens, priceData.CacheRatio, priceData.ModelPrice, priceData.UserGroupRatio)
+	other := service.GenerateTextOtherInfo(c, info, priceData.ModelRatio, priceData.GroupRatioInfo.GroupRatio, priceData.CompletionRatio,
+		usage.PromptTokensDetails.CachedTokens, priceData.CacheRatio, priceData.ModelPrice, priceData.GroupRatioInfo.GroupSpecialRatio)
 	model.RecordConsumeLog(c, 1, channel.Id, usage.PromptTokens, usage.CompletionTokens, info.OriginModelName, "模型测试",
 		quota, "模型测试", 0, quota, int(consumedTime), false, info.Group, other)
 	common.SysLog(fmt.Sprintf("testing channel #%d, response: \n%s", channel.Id, string(respBody)))
@@ -312,7 +312,7 @@ func testAllChannels(notify bool) error {
 			channel.UpdateResponseTime(milliseconds)
 			time.Sleep(common.RequestInterval)
 		}
-		
+
 		if notify {
 			service.NotifyRootUser(dto.NotifyTypeChannelTest, "通道测试完成", "所有通道测试已完成")
 		}

controller/group.go

@@ -1,10 +1,11 @@
 package controller
 
 import (
-	"github.com/gin-gonic/gin"
 	"net/http"
 	"one-api/model"
 	"one-api/setting"
+
+	"github.com/gin-gonic/gin"
 )
 
 func GetGroups(c *gin.Context) {
@@ -34,6 +35,12 @@ func GetUserGroups(c *gin.Context) {
 			}
 		}
 	}
+	if setting.GroupInUserUsableGroups("auto") {
+		usableGroups["auto"] = map[string]interface{}{
+			"ratio": "自动",
+			"desc":  setting.GetUsableGroupDescription("auto"),
+		}
+	}
 	c.JSON(http.StatusOK, gin.H{
 		"success": true,
 		"message": "",

controller/misc.go

@@ -9,9 +9,9 @@ import (
 	"one-api/middleware"
 	"one-api/model"
 	"one-api/setting"
+	"one-api/setting/console_setting"
 	"one-api/setting/operation_setting"
 	"one-api/setting/system_setting"
-	"one-api/setting/console_setting"
 	"strings"
 
 	"github.com/gin-gonic/gin"
@@ -41,46 +41,47 @@ func GetStatus(c *gin.Context) {
 	cs := console_setting.GetConsoleSetting()
 
 	data := gin.H{
-		"version":                     common.Version,
-		"start_time":                  common.StartTime,
-		"email_verification":          common.EmailVerificationEnabled,
-		"github_oauth":                common.GitHubOAuthEnabled,
-		"github_client_id":            common.GitHubClientId,
-		"linuxdo_oauth":               common.LinuxDOOAuthEnabled,
-		"linuxdo_client_id":           common.LinuxDOClientId,
-		"telegram_oauth":              common.TelegramOAuthEnabled,
-		"telegram_bot_name":           common.TelegramBotName,
-		"system_name":                 common.SystemName,
-		"logo":                        common.Logo,
-		"footer_html":                 common.Footer,
-		"wechat_qrcode":               common.WeChatAccountQRCodeImageURL,
-		"wechat_login":                common.WeChatAuthEnabled,
-		"server_address":              setting.ServerAddress,
-		"price":                       setting.Price,
-		"min_topup":                   setting.MinTopUp,
-		"turnstile_check":             common.TurnstileCheckEnabled,
-		"turnstile_site_key":          common.TurnstileSiteKey,
-		"top_up_link":                 common.TopUpLink,
-		"docs_link":                   operation_setting.GetGeneralSetting().DocsLink,
-		"quota_per_unit":              common.QuotaPerUnit,
-		"display_in_currency":         common.DisplayInCurrencyEnabled,
-		"enable_batch_update":         common.BatchUpdateEnabled,
-		"enable_drawing":              common.DrawingEnabled,
-		"enable_task":                 common.TaskEnabled,
-		"enable_data_export":          common.DataExportEnabled,
-		"data_export_default_time":    common.DataExportDefaultTime,
-		"default_collapse_sidebar":    common.DefaultCollapseSidebar,
-		"enable_online_topup":         setting.PayAddress != "" && setting.EpayId != "" && setting.EpayKey != "",
-		"mj_notify_enabled":           setting.MjNotifyEnabled,
-		"chats":                       setting.Chats,
-		"demo_site_enabled":           operation_setting.DemoSiteEnabled,
-		"self_use_mode_enabled":       operation_setting.SelfUseModeEnabled,
+		"version":                  common.Version,
+		"start_time":               common.StartTime,
+		"email_verification":       common.EmailVerificationEnabled,
+		"github_oauth":             common.GitHubOAuthEnabled,
+		"github_client_id":         common.GitHubClientId,
+		"linuxdo_oauth":            common.LinuxDOOAuthEnabled,
+		"linuxdo_client_id":        common.LinuxDOClientId,
+		"telegram_oauth":           common.TelegramOAuthEnabled,
+		"telegram_bot_name":        common.TelegramBotName,
+		"system_name":              common.SystemName,
+		"logo":                     common.Logo,
+		"footer_html":              common.Footer,
+		"wechat_qrcode":            common.WeChatAccountQRCodeImageURL,
+		"wechat_login":             common.WeChatAuthEnabled,
+		"server_address":           setting.ServerAddress,
+		"price":                    setting.Price,
+		"min_topup":                setting.MinTopUp,
+		"turnstile_check":          common.TurnstileCheckEnabled,
+		"turnstile_site_key":       common.TurnstileSiteKey,
+		"top_up_link":              common.TopUpLink,
+		"docs_link":                operation_setting.GetGeneralSetting().DocsLink,
+		"quota_per_unit":           common.QuotaPerUnit,
+		"display_in_currency":      common.DisplayInCurrencyEnabled,
+		"enable_batch_update":      common.BatchUpdateEnabled,
+		"enable_drawing":           common.DrawingEnabled,
+		"enable_task":              common.TaskEnabled,
+		"enable_data_export":       common.DataExportEnabled,
+		"data_export_default_time": common.DataExportDefaultTime,
+		"default_collapse_sidebar": common.DefaultCollapseSidebar,
+		"enable_online_topup":      setting.PayAddress != "" && setting.EpayId != "" && setting.EpayKey != "",
+		"mj_notify_enabled":        setting.MjNotifyEnabled,
+		"chats":                    setting.Chats,
+		"demo_site_enabled":        operation_setting.DemoSiteEnabled,
+		"self_use_mode_enabled":    operation_setting.SelfUseModeEnabled,
+		"default_use_auto_group":   setting.DefaultUseAutoGroup,
 
 		// 面板启用开关
-		"api_info_enabled":            cs.ApiInfoEnabled,
-		"uptime_kuma_enabled":         cs.UptimeKumaEnabled,
-		"announcements_enabled":       cs.AnnouncementsEnabled,
-		"faq_enabled":                 cs.FAQEnabled,
+		"api_info_enabled":      cs.ApiInfoEnabled,
+		"uptime_kuma_enabled":   cs.UptimeKumaEnabled,
+		"announcements_enabled": cs.AnnouncementsEnabled,
+		"faq_enabled":           cs.FAQEnabled,
 
 		"oidc_enabled":                system_setting.GetOIDCSettings().Enabled,
 		"oidc_client_id":              system_setting.GetOIDCSettings().ClientId,

controller/model.go

@@ -2,7 +2,6 @@ package controller
 
 import (
 	"fmt"
-	"github.com/gin-gonic/gin"
 	"net/http"
 	"one-api/common"
 	"one-api/constant"
@@ -15,6 +14,9 @@ import (
 	"one-api/relay/channel/moonshot"
 	relaycommon "one-api/relay/common"
 	relayconstant "one-api/relay/constant"
+	"one-api/setting"
+
+	"github.com/gin-gonic/gin"
 )
 
 // https://platform.openai.com/docs/api-reference/models/list
@@ -179,7 +181,19 @@ func ListModels(c *gin.Context) {
 		if tokenGroup != "" {
 			group = tokenGroup
 		}
-		models := model.GetGroupModels(group)
+		var models []string
+		if tokenGroup == "auto" {
+			for _, autoGroup := range setting.AutoGroups {
+				groupModels := model.GetGroupModels(autoGroup)
+				for _, g := range groupModels {
+					if !common.StringsContains(models, g) {
+						models = append(models, g)
+					}
+				}
+			}
+		} else {
+			models = model.GetGroupModels(group)
+		}
 		for _, s := range models {
 			if _, ok := openAIModelsMap[s]; ok {
 				userOpenAiModels = append(userOpenAiModels, openAIModelsMap[s])

controller/playground.go

@@ -3,7 +3,6 @@ package controller
 import (
 	"errors"
 	"fmt"
-	"github.com/gin-gonic/gin"
 	"net/http"
 	"one-api/common"
 	"one-api/constant"
@@ -13,6 +12,8 @@ import (
 	"one-api/service"
 	"one-api/setting"
 	"time"
+
+	"github.com/gin-gonic/gin"
 )
 
 func Playground(c *gin.Context) {
@@ -57,9 +58,9 @@ func Playground(c *gin.Context) {
 		c.Set("group", group)
 	}
 	c.Set("token_name", "playground-"+group)
-	channel, err := model.CacheGetRandomSatisfiedChannel(group, playgroundRequest.Model, 0)
+	channel, finalGroup, err := model.CacheGetRandomSatisfiedChannel(c, group, playgroundRequest.Model, 0)
 	if err != nil {
-		message := fmt.Sprintf("当前分组 %s 下对于模型 %s 无可用渠道", group, playgroundRequest.Model)
+		message := fmt.Sprintf("当前分组 %s 下对于模型 %s 无可用渠道", finalGroup, playgroundRequest.Model)
 		openaiErr = service.OpenAIErrorWrapperLocal(errors.New(message), "get_playground_channel_failed", http.StatusInternalServerError)
 		return
 	}

controller/relay.go

@@ -259,7 +259,7 @@ func getChannel(c *gin.Context, group, originalModel string, retryCount int) (*m
 			AutoBan: &autoBanInt,
 		}, nil
 	}
-	channel, err := model.CacheGetRandomSatisfiedChannel(group, originalModel, retryCount)
+	channel, _, err := model.CacheGetRandomSatisfiedChannel(c, group, originalModel, retryCount)
 	if err != nil {
 		return nil, errors.New(fmt.Sprintf("获取重试渠道失败: %s", err.Error()))
 	}
@@ -388,7 +388,7 @@ func RelayTask(c *gin.Context) {
 		retryTimes = 0
 	}
 	for i := 0; shouldRetryTaskRelay(c, channelId, taskErr, retryTimes) && i < retryTimes; i++ {
-		channel, err := model.CacheGetRandomSatisfiedChannel(group, originalModel, i)
+		channel, _, err := model.CacheGetRandomSatisfiedChannel(c, group, originalModel, i)
 		if err != nil {
 			common.LogError(c, fmt.Sprintf("CacheGetRandomSatisfiedChannel failed: %s", err.Error()))
 			break

controller/user.go

@@ -226,6 +226,9 @@ func Register(c *gin.Context) {
 			UnlimitedQuota:     true,
 			ModelLimitsEnabled: false,
 		}
+		if setting.DefaultUseAutoGroup {
+			token.Group = "auto"
+		}
 		if err := token.Insert(); err != nil {
 			c.JSON(http.StatusOK, gin.H{
 				"success": false,

main.go

@@ -105,10 +105,12 @@ func main() {
 			model.InitChannelCache()
 		}()
 
-		go model.SyncOptions(common.SyncFrequency)
 		go model.SyncChannelCache(common.SyncFrequency)
 	}
 
+	// 热更新配置
+	go model.SyncOptions(common.SyncFrequency)
+
 	// 数据看板
 	go model.UpdateQuotaData()
 

middleware/distributor.go

@@ -49,8 +49,10 @@ func Distribute() func(c *gin.Context) {
 			}
 			// check group in common.GroupRatio
 			if !setting.ContainsGroupRatio(tokenGroup) {
-				abortWithOpenAiMessage(c, http.StatusForbidden, fmt.Sprintf("分组 %s 已被弃用", tokenGroup))
-				return
+				if tokenGroup != "auto" {
+					abortWithOpenAiMessage(c, http.StatusForbidden, fmt.Sprintf("分组 %s 已被弃用", tokenGroup))
+					return
+				}
 			}
 			userGroup = tokenGroup
 		}
@@ -95,9 +97,14 @@ func Distribute() func(c *gin.Context) {
 			}
 
 			if shouldSelectChannel {
-				channel, err = model.CacheGetRandomSatisfiedChannel(userGroup, modelRequest.Model, 0)
+				var selectGroup string
+				channel, selectGroup, err = model.CacheGetRandomSatisfiedChannel(c, userGroup, modelRequest.Model, 0)
 				if err != nil {
-					message := fmt.Sprintf("当前分组 %s 下对于模型 %s 无可用渠道", userGroup, modelRequest.Model)
+					showGroup := userGroup
+					if userGroup == "auto" {
+						showGroup = fmt.Sprintf("auto(%s)", selectGroup)
+					}
+					message := fmt.Sprintf("当前分组 %s 下对于模型 %s 无可用渠道", showGroup, modelRequest.Model)
 					// 如果错误，但是渠道不为空，说明是数据库一致性问题
 					if channel != nil {
 						common.SysError(fmt.Sprintf("渠道不存在：%d", channel.Id))

model/cache.go

@@ -5,10 +5,13 @@ import (
 	"fmt"
 	"math/rand"
 	"one-api/common"
+	"one-api/setting"
 	"sort"
 	"strings"
 	"sync"
 	"time"
+
+	"github.com/gin-gonic/gin"
 )
 
 var group2model2channels map[string]map[string][]*Channel
@@ -75,7 +78,43 @@ func SyncChannelCache(frequency int) {
 	}
 }
 
-func CacheGetRandomSatisfiedChannel(group string, model string, retry int) (*Channel, error) {
+func CacheGetRandomSatisfiedChannel(c *gin.Context, group string, model string, retry int) (*Channel, string, error) {
+	var channel *Channel
+	var err error
+	selectGroup := group
+	if group == "auto" {
+		if len(setting.AutoGroups) == 0 {
+			return nil, selectGroup, errors.New("auto groups is not enabled")
+		}
+		for _, autoGroup := range setting.AutoGroups {
+			if common.DebugEnabled {
+				println("autoGroup:", autoGroup)
+			}
+			channel, _ = getRandomSatisfiedChannel(autoGroup, model, retry)
+			if channel == nil {
+				continue
+			} else {
+				c.Set("auto_group", autoGroup)
+				selectGroup = autoGroup
+				if common.DebugEnabled {
+					println("selectGroup:", selectGroup)
+				}
+				break
+			}
+		}
+	} else {
+		channel, err = getRandomSatisfiedChannel(group, model, retry)
+		if err != nil {
+			return nil, group, err
+		}
+	}
+	if channel == nil {
+		return nil, group, errors.New("channel not found")
+	}
+	return channel, selectGroup, nil
+}
+
+func getRandomSatisfiedChannel(group string, model string, retry int) (*Channel, error) {
 	if strings.HasPrefix(model, "gpt-4-gizmo") {
 		model = "gpt-4-gizmo-*"
 	}

model/option.go

@@ -76,6 +76,8 @@ func InitOptionMap() {
 	common.OptionMap["MinTopUp"] = strconv.Itoa(setting.MinTopUp)
 	common.OptionMap["TopupGroupRatio"] = common.TopupGroupRatio2JSONString()
 	common.OptionMap["Chats"] = setting.Chats2JsonString()
+	common.OptionMap["AutoGroups"] = setting.AutoGroups2JsonString()
+	common.OptionMap["DefaultUseAutoGroup"] = strconv.FormatBool(setting.DefaultUseAutoGroup)
 	common.OptionMap["GitHubClientId"] = ""
 	common.OptionMap["GitHubClientSecret"] = ""
 	common.OptionMap["TelegramBotToken"] = ""
@@ -192,7 +194,7 @@ func updateOptionMap(key string, value string) (err error) {
 			common.ImageDownloadPermission = intValue
 		}
 	}
-	if strings.HasSuffix(key, "Enabled") || key == "DefaultCollapseSidebar" {
+	if strings.HasSuffix(key, "Enabled") || key == "DefaultCollapseSidebar" || key == "DefaultUseAutoGroup" {
 		boolValue := value == "true"
 		switch key {
 		case "PasswordRegisterEnabled":
@@ -261,6 +263,8 @@ func updateOptionMap(key string, value string) (err error) {
 			common.SMTPSSLEnabled = boolValue
 		case "WorkerAllowHttpImageRequestEnabled":
 			setting.WorkerAllowHttpImageRequestEnabled = boolValue
+		case "DefaultUseAutoGroup":
+			setting.DefaultUseAutoGroup = boolValue
 		}
 	}
 	switch key {
@@ -287,6 +291,8 @@ func updateOptionMap(key string, value string) (err error) {
 		setting.PayAddress = value
 	case "Chats":
 		err = setting.UpdateChatsByJsonString(value)
+	case "AutoGroups":
+		err = setting.UpdateAutoGroupsByJsonString(value)
 	case "CustomCallbackAddress":
 		setting.CustomCallbackAddress = value
 	case "EpayId":

model/token_cache.go

@@ -10,7 +10,7 @@ import (
 func cacheSetToken(token Token) error {
 	key := common.GenerateHMAC(token.Key)
 	token.Clean()
-	err := common.RedisHSetObj(fmt.Sprintf("token:%s", key), &token, time.Duration(constant.TokenCacheSeconds)*time.Second)
+	err := common.RedisHSetObj(fmt.Sprintf("token:%s", key), &token, time.Duration(constant.RedisKeyCacheSeconds())*time.Second)
 	if err != nil {
 		return err
 	}

model/user_cache.go

@@ -70,7 +70,7 @@ func updateUserCache(user User) error {
 	return common.RedisHSetObj(
 		getUserCacheKey(user.Id),
 		user.ToBaseUser(),
-		time.Duration(constant.UserId2QuotaCacheSeconds)*time.Second,
+		time.Duration(constant.RedisKeyCacheSeconds())*time.Second,
 	)
 }
 

relay/channel/claude/relay-claude.go

@@ -454,6 +454,7 @@ type ClaudeResponseInfo struct {
 	Model        string
 	ResponseText strings.Builder
 	Usage        *dto.Usage
+	Done         bool
 }
 
 func FormatClaudeResponseInfo(requestMode int, claudeResponse *dto.ClaudeResponse, oaiResponse *dto.ChatCompletionsStreamResponse, claudeInfo *ClaudeResponseInfo) bool {
@@ -461,20 +462,32 @@ func FormatClaudeResponseInfo(requestMode int, claudeResponse *dto.ClaudeRespons
 		claudeInfo.ResponseText.WriteString(claudeResponse.Completion)
 	} else {
 		if claudeResponse.Type == "message_start" {
-			// message_start, 获取usage
 			claudeInfo.ResponseId = claudeResponse.Message.Id
 			claudeInfo.Model = claudeResponse.Message.Model
+
+			// message_start, 获取usage
 			claudeInfo.Usage.PromptTokens = claudeResponse.Message.Usage.InputTokens
+			claudeInfo.Usage.PromptTokensDetails.CachedTokens = claudeResponse.Message.Usage.CacheReadInputTokens
+			claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens = claudeResponse.Message.Usage.CacheCreationInputTokens
+			claudeInfo.Usage.CompletionTokens = claudeResponse.Message.Usage.OutputTokens
 		} else if claudeResponse.Type == "content_block_delta" {
 			if claudeResponse.Delta.Text != nil {
 				claudeInfo.ResponseText.WriteString(*claudeResponse.Delta.Text)
 			}
+			if claudeResponse.Delta.Thinking != "" {
+				claudeInfo.ResponseText.WriteString(claudeResponse.Delta.Thinking)
+			}
 		} else if claudeResponse.Type == "message_delta" {
-			claudeInfo.Usage.CompletionTokens = claudeResponse.Usage.OutputTokens
+			// 最终的usage获取
 			if claudeResponse.Usage.InputTokens > 0 {
+				// 不叠加，只取最新的
 				claudeInfo.Usage.PromptTokens = claudeResponse.Usage.InputTokens
 			}
-			claudeInfo.Usage.TotalTokens = claudeInfo.Usage.PromptTokens + claudeResponse.Usage.OutputTokens
+			claudeInfo.Usage.CompletionTokens = claudeResponse.Usage.OutputTokens
+			claudeInfo.Usage.TotalTokens = claudeInfo.Usage.PromptTokens + claudeInfo.Usage.CompletionTokens
+
+			// 判断是否完整
+			claudeInfo.Done = true
 		} else if claudeResponse.Type == "content_block_start" {
 		} else {
 			return false
@@ -506,25 +519,15 @@ func HandleStreamResponseData(c *gin.Context, info *relaycommon.RelayInfo, claud
 		}
 	}
 	if info.RelayFormat == relaycommon.RelayFormatClaude {
+		FormatClaudeResponseInfo(requestMode, &claudeResponse, nil, claudeInfo)
+
 		if requestMode == RequestModeCompletion {
-			claudeInfo.ResponseText.WriteString(claudeResponse.Completion)
 		} else {
 			if claudeResponse.Type == "message_start" {
 				// message_start, 获取usage
 				info.UpstreamModelName = claudeResponse.Message.Model
-				claudeInfo.Usage.PromptTokens = claudeResponse.Message.Usage.InputTokens
-				claudeInfo.Usage.PromptTokensDetails.CachedTokens = claudeResponse.Message.Usage.CacheReadInputTokens
-				claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens = claudeResponse.Message.Usage.CacheCreationInputTokens
-				claudeInfo.Usage.CompletionTokens = claudeResponse.Message.Usage.OutputTokens
 			} else if claudeResponse.Type == "content_block_delta" {
-				claudeInfo.ResponseText.WriteString(claudeResponse.Delta.GetText())
 			} else if claudeResponse.Type == "message_delta" {
-				if claudeResponse.Usage.InputTokens > 0 {
-					// 不叠加，只取最新的
-					claudeInfo.Usage.PromptTokens = claudeResponse.Usage.InputTokens
-				}
-				claudeInfo.Usage.CompletionTokens = claudeResponse.Usage.OutputTokens
-				claudeInfo.Usage.TotalTokens = claudeInfo.Usage.PromptTokens + claudeInfo.Usage.CompletionTokens
 			}
 		}
 		helper.ClaudeChunkData(c, claudeResponse, data)
@@ -544,29 +547,25 @@ func HandleStreamResponseData(c *gin.Context, info *relaycommon.RelayInfo, claud
 }
 
 func HandleStreamFinalResponse(c *gin.Context, info *relaycommon.RelayInfo, claudeInfo *ClaudeResponseInfo, requestMode int) {
+
+	if requestMode == RequestModeCompletion {
+		claudeInfo.Usage, _ = service.ResponseText2Usage(claudeInfo.ResponseText.String(), info.UpstreamModelName, info.PromptTokens)
+	} else {
+		if claudeInfo.Usage.PromptTokens == 0 {
+			//上游出错
+		}
+		if claudeInfo.Usage.CompletionTokens == 0 || !claudeInfo.Done {
+			if common.DebugEnabled {
+				common.SysError("claude response usage is not complete, maybe upstream error")
+			}
+			claudeInfo.Usage, _ = service.ResponseText2Usage(claudeInfo.ResponseText.String(), info.UpstreamModelName, claudeInfo.Usage.PromptTokens)
+		}
+	}
+
 	if info.RelayFormat == relaycommon.RelayFormatClaude {
-		if requestMode == RequestModeCompletion {
-			claudeInfo.Usage, _ = service.ResponseText2Usage(claudeInfo.ResponseText.String(), info.UpstreamModelName, info.PromptTokens)
-		} else {
-			// 说明流模式建立失败，可能为官方出错
-			if claudeInfo.Usage.PromptTokens == 0 {
-				//usage.PromptTokens = info.PromptTokens
-			}
-			if claudeInfo.Usage.CompletionTokens == 0 {
-				claudeInfo.Usage, _ = service.ResponseText2Usage(claudeInfo.ResponseText.String(), info.UpstreamModelName, claudeInfo.Usage.PromptTokens)
-			}
-		}
+		//
 	} else if info.RelayFormat == relaycommon.RelayFormatOpenAI {
-		if requestMode == RequestModeCompletion {
-			claudeInfo.Usage, _ = service.ResponseText2Usage(claudeInfo.ResponseText.String(), info.UpstreamModelName, info.PromptTokens)
-		} else {
-			if claudeInfo.Usage.PromptTokens == 0 {
-				//上游出错
-			}
-			if claudeInfo.Usage.CompletionTokens == 0 {
-				claudeInfo.Usage, _ = service.ResponseText2Usage(claudeInfo.ResponseText.String(), info.UpstreamModelName, claudeInfo.Usage.PromptTokens)
-			}
-		}
+
 		if info.ShouldIncludeUsage {
 			response := helper.GenerateFinalUsageResponse(claudeInfo.ResponseId, claudeInfo.Created, info.UpstreamModelName, *claudeInfo.Usage)
 			err := helper.ObjectData(c, response)

relay/channel/cohere/relay-cohere.go

@@ -3,7 +3,6 @@ package cohere
 import (
 	"bufio"
 	"encoding/json"
-	"fmt"
 	"github.com/gin-gonic/gin"
 	"io"
 	"net/http"
@@ -78,7 +77,7 @@ func stopReasonCohere2OpenAI(reason string) string {
 }
 
 func cohereStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.OpenAIErrorWithStatusCode, *dto.Usage) {
-	responseId := fmt.Sprintf("chatcmpl-%s", common.GetUUID())
+	responseId := helper.GetResponseID(c)
 	createdTime := common.GetTimestamp()
 	usage := &dto.Usage{}
 	responseText := ""

relay/channel/gemini/adaptor.go

@@ -72,8 +72,11 @@ func (a *Adaptor) Init(info *relaycommon.RelayInfo) {
 func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
 
 	if model_setting.GetGeminiSettings().ThinkingAdapterEnabled {
-		// suffix -thinking and -nothinking
-		if strings.HasSuffix(info.OriginModelName, "-thinking") {
+		// 新增逻辑：处理 -thinking-<budget> 格式
+		if strings.Contains(info.OriginModelName, "-thinking-") {
+			parts := strings.Split(info.UpstreamModelName, "-thinking-")
+			info.UpstreamModelName = parts[0]
+		} else if strings.HasSuffix(info.OriginModelName, "-thinking") { // 旧的适配
 			info.UpstreamModelName = strings.TrimSuffix(info.UpstreamModelName, "-thinking")
 		} else if strings.HasSuffix(info.OriginModelName, "-nothinking") {
 			info.UpstreamModelName = strings.TrimSuffix(info.UpstreamModelName, "-nothinking")

relay/channel/gemini/relay-gemini.go

@@ -12,6 +12,7 @@ import (
 	"one-api/relay/helper"
 	"one-api/service"
 	"one-api/setting/model_setting"
+	"strconv"
 	"strings"
 	"unicode/utf8"
 
@@ -36,6 +37,47 @@ var geminiSupportedMimeTypes = map[string]bool{
 	"video/flv":       true,
 }
 
+// Gemini 允许的思考预算范围
+const (
+	pro25MinBudget       = 128
+	pro25MaxBudget       = 32768
+	flash25MaxBudget     = 24576
+	flash25LiteMinBudget = 512
+	flash25LiteMaxBudget = 24576
+)
+
+// clampThinkingBudget 根据模型名称将预算限制在允许的范围内
+func clampThinkingBudget(modelName string, budget int) int {
+	isNew25Pro := strings.HasPrefix(modelName, "gemini-2.5-pro") &&
+		!strings.HasPrefix(modelName, "gemini-2.5-pro-preview-05-06") &&
+		!strings.HasPrefix(modelName, "gemini-2.5-pro-preview-03-25")
+	is25FlashLite := strings.HasPrefix(modelName, "gemini-2.5-flash-lite")
+
+	if is25FlashLite {
+		if budget < flash25LiteMinBudget {
+			return flash25LiteMinBudget
+		}
+		if budget > flash25LiteMaxBudget {
+			return flash25LiteMaxBudget
+		}
+	} else if isNew25Pro {
+		if budget < pro25MinBudget {
+			return pro25MinBudget
+		}
+		if budget > pro25MaxBudget {
+			return pro25MaxBudget
+		}
+	} else { // 其他模型
+		if budget < 0 {
+			return 0
+		}
+		if budget > flash25MaxBudget {
+			return flash25MaxBudget
+		}
+	}
+	return budget
+}
+
 // Setting safety to the lowest possible values since Gemini is already powerless enough
 func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon.RelayInfo) (*GeminiChatRequest, error) {
 
@@ -57,16 +99,31 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 	}
 
 	if model_setting.GetGeminiSettings().ThinkingAdapterEnabled {
-		if strings.HasSuffix(info.OriginModelName, "-thinking") {
-			// 硬编码不支持 ThinkingBudget 的旧模型
+		modelName := info.OriginModelName
+		isNew25Pro := strings.HasPrefix(modelName, "gemini-2.5-pro") &&
+			!strings.HasPrefix(modelName, "gemini-2.5-pro-preview-05-06") &&
+			!strings.HasPrefix(modelName, "gemini-2.5-pro-preview-03-25")
+		is25FlashLite := strings.HasPrefix(modelName, "gemini-2.5-flash-lite")
+
+		if strings.Contains(modelName, "-thinking-") {
+			parts := strings.SplitN(modelName, "-thinking-", 2)
+			if len(parts) == 2 && parts[1] != "" {
+				if budgetTokens, err := strconv.Atoi(parts[1]); err == nil {
+					clampedBudget := clampThinkingBudget(modelName, budgetTokens)
+					geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
+						ThinkingBudget:  common.GetPointer(clampedBudget),
+						IncludeThoughts: true,
+					}
+				}
+			}
+		} else if strings.HasSuffix(modelName, "-thinking") {
 			unsupportedModels := []string{
 				"gemini-2.5-pro-preview-05-06",
 				"gemini-2.5-pro-preview-03-25",
 			}
-
 			isUnsupported := false
 			for _, unsupportedModel := range unsupportedModels {
-				if strings.HasPrefix(info.OriginModelName, unsupportedModel) {
+				if strings.HasPrefix(modelName, unsupportedModel) {
 					isUnsupported = true
 					break
 				}
@@ -78,39 +135,14 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 				}
 			} else {
 				budgetTokens := model_setting.GetGeminiSettings().ThinkingAdapterBudgetTokensPercentage * float64(geminiRequest.GenerationConfig.MaxOutputTokens)
-
-				// 检查是否为新的2.5pro模型（支持ThinkingBudget但有特殊范围）
-				isNew25Pro := strings.HasPrefix(info.OriginModelName, "gemini-2.5-pro") &&
-					!strings.HasPrefix(info.OriginModelName, "gemini-2.5-pro-preview-05-06") &&
-					!strings.HasPrefix(info.OriginModelName, "gemini-2.5-pro-preview-03-25")
-
-				if isNew25Pro {
-					// 新的2.5pro模型：ThinkingBudget范围为128-32768
-					if budgetTokens == 0 || budgetTokens < 128 {
-						budgetTokens = 128
-					} else if budgetTokens > 32768 {
-						budgetTokens = 32768
-					}
-				} else {
-					// 其他模型：ThinkingBudget范围为0-24576
-					if budgetTokens == 0 || budgetTokens > 24576 {
-						budgetTokens = 24576
-					}
-				}
-
+				clampedBudget := clampThinkingBudget(modelName, int(budgetTokens))
 				geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
-					ThinkingBudget:  common.GetPointer(int(budgetTokens)),
+					ThinkingBudget:  common.GetPointer(clampedBudget),
 					IncludeThoughts: true,
 				}
 			}
-		} else if strings.HasSuffix(info.OriginModelName, "-nothinking") {
-			// 检查是否为新的2.5pro模型（不支持-nothinking，因为最低值只能为128）
-			isNew25Pro := strings.HasPrefix(info.OriginModelName, "gemini-2.5-pro") &&
-				!strings.HasPrefix(info.OriginModelName, "gemini-2.5-pro-preview-05-06") &&
-				!strings.HasPrefix(info.OriginModelName, "gemini-2.5-pro-preview-03-25")
-
-			if !isNew25Pro {
-				// 只有非新2.5pro模型才支持-nothinking
+		} else if strings.HasSuffix(modelName, "-nothinking") {
+			if !isNew25Pro && !is25FlashLite {
 				geminiRequest.GenerationConfig.ThinkingConfig = &GeminiThinkingConfig{
 					ThinkingBudget: common.GetPointer(0),
 				}
@@ -283,7 +315,8 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 
 					// 校验 MimeType 是否在 Gemini 支持的白名单中
 					if _, ok := geminiSupportedMimeTypes[strings.ToLower(fileData.MimeType)]; !ok {
-						return nil, fmt.Errorf("MIME type '%s' from URL '%s' is not supported by Gemini. Supported types are: %v", fileData.MimeType, part.GetImageMedia().Url, getSupportedMimeTypesList())
+						url := part.GetImageMedia().Url
+						return nil, fmt.Errorf("mime type is not supported by Gemini: '%s', url: '%s', supported types are: %v", fileData.MimeType, url, getSupportedMimeTypesList())
 					}
 
 					parts = append(parts, GeminiPart{
@@ -611,9 +644,9 @@ func getResponseToolCall(item *GeminiPart) *dto.ToolCallResponse {
 	}
 }
 
-func responseGeminiChat2OpenAI(response *GeminiChatResponse) *dto.OpenAITextResponse {
+func responseGeminiChat2OpenAI(c *gin.Context, response *GeminiChatResponse) *dto.OpenAITextResponse {
 	fullTextResponse := dto.OpenAITextResponse{
-		Id:      fmt.Sprintf("chatcmpl-%s", common.GetUUID()),
+		Id:      helper.GetResponseID(c),
 		Object:  "chat.completion",
 		Created: common.GetTimestamp(),
 		Choices: make([]dto.OpenAITextResponseChoice, 0, len(response.Candidates)),
@@ -754,7 +787,7 @@ func streamResponseGeminiChat2OpenAI(geminiResponse *GeminiChatResponse) (*dto.C
 
 func GeminiChatStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.OpenAIErrorWithStatusCode, *dto.Usage) {
 	// responseText := ""
-	id := fmt.Sprintf("chatcmpl-%s", common.GetUUID())
+	id := helper.GetResponseID(c)
 	createAt := common.GetTimestamp()
 	var usage = &dto.Usage{}
 	var imageCount int
@@ -849,7 +882,7 @@ func GeminiChatHandler(c *gin.Context, resp *http.Response, info *relaycommon.Re
 			StatusCode: resp.StatusCode,
 		}, nil
 	}
-	fullTextResponse := responseGeminiChat2OpenAI(&geminiResponse)
+	fullTextResponse := responseGeminiChat2OpenAI(c, &geminiResponse)
 	fullTextResponse.Model = info.UpstreamModelName
 	usage := dto.Usage{
 		PromptTokens:     geminiResponse.UsageMetadata.PromptTokenCount,

relay/channel/openai/adaptor.go

@@ -88,6 +88,13 @@ func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
 		requestURL := strings.Split(info.RequestURLPath, "?")[0]
 		requestURL = fmt.Sprintf("%s?api-version=%s", requestURL, apiVersion)
 		task := strings.TrimPrefix(requestURL, "/v1/")
+
+		// 特殊处理 responses API
+		if info.RelayMode == constant.RelayModeResponses {
+			requestURL = fmt.Sprintf("/openai/v1/responses?api-version=preview")
+			return relaycommon.GetFullRequestURL(info.BaseUrl, requestURL, info.ChannelType), nil
+		}
+
 		model_ := info.UpstreamModelName
 		// 2025年5月10日后创建的渠道不移除.
 		if info.ChannelCreateTime < constant2.AzureNoRemoveDotTime {

relay/channel/openai/relay-openai.go

@@ -8,6 +8,7 @@ import (
 	"math"
 	"mime/multipart"
 	"net/http"
+	"path/filepath"
 	"one-api/common"
 	"one-api/constant"
 	"one-api/dto"
@@ -345,13 +346,14 @@ func countAudioTokens(c *gin.Context) (int, error) {
 	if err = c.ShouldBind(&reqBody); err != nil {
 		return 0, errors.WithStack(err)
 	}
-
+  ext := filepath.Ext(reqBody.File.Filename) // 获取文件扩展名
 	reqFp, err := reqBody.File.Open()
 	if err != nil {
 		return 0, errors.WithStack(err)
 	}
+  defer reqFp.Close()
 
-	tmpFp, err := os.CreateTemp("", "audio-*")
+	tmpFp, err := os.CreateTemp("", "audio-*"+ext)
 	if err != nil {
 		return 0, errors.WithStack(err)
 	}
@@ -365,7 +367,7 @@ func countAudioTokens(c *gin.Context) (int, error) {
 		return 0, errors.WithStack(err)
 	}
 
-	duration, err := common.GetAudioDuration(c.Request.Context(), tmpFp.Name())
+	duration, err := common.GetAudioDuration(c.Request.Context(), tmpFp.Name(), ext)
 	if err != nil {
 		return 0, errors.WithStack(err)
 	}

relay/channel/palm/relay-palm.go

@@ -2,7 +2,6 @@ package palm
 
 import (
 	"encoding/json"
-	"fmt"
 	"github.com/gin-gonic/gin"
 	"io"
 	"net/http"
@@ -73,7 +72,7 @@ func streamResponsePaLM2OpenAI(palmResponse *PaLMChatResponse) *dto.ChatCompleti
 
 func palmStreamHandler(c *gin.Context, resp *http.Response) (*dto.OpenAIErrorWithStatusCode, string) {
 	responseText := ""
-	responseId := fmt.Sprintf("chatcmpl-%s", common.GetUUID())
+	responseId := helper.GetResponseID(c)
 	createdTime := common.GetTimestamp()
 	dataChan := make(chan string)
 	stopChan := make(chan bool)

relay/helper/price.go

@@ -11,6 +11,11 @@ import (
 	"github.com/gin-gonic/gin"
 )
 
+type GroupRatioInfo struct {
+	GroupRatio        float64
+	GroupSpecialRatio float64
+}
+
 type PriceData struct {
 	ModelPrice             float64
 	ModelRatio             float64
@@ -18,23 +23,50 @@ type PriceData struct {
 	CacheRatio             float64
 	CacheCreationRatio     float64
 	ImageRatio             float64
-	GroupRatio             float64
-	UserGroupRatio         float64
 	UsePrice               bool
 	ShouldPreConsumedQuota int
+	GroupRatioInfo         GroupRatioInfo
 }
 
 func (p PriceData) ToSetting() string {
-	return fmt.Sprintf("ModelPrice: %f, ModelRatio: %f, CompletionRatio: %f, CacheRatio: %f, GroupRatio: %f, UsePrice: %t, CacheCreationRatio: %f, ShouldPreConsumedQuota: %d, ImageRatio: %f", p.ModelPrice, p.ModelRatio, p.CompletionRatio, p.CacheRatio, p.GroupRatio, p.UsePrice, p.CacheCreationRatio, p.ShouldPreConsumedQuota, p.ImageRatio)
+	return fmt.Sprintf("ModelPrice: %f, ModelRatio: %f, CompletionRatio: %f, CacheRatio: %f, GroupRatio: %f, UsePrice: %t, CacheCreationRatio: %f, ShouldPreConsumedQuota: %d, ImageRatio: %f", p.ModelPrice, p.ModelRatio, p.CompletionRatio, p.CacheRatio, p.GroupRatioInfo.GroupRatio, p.UsePrice, p.CacheCreationRatio, p.ShouldPreConsumedQuota, p.ImageRatio)
+}
+
+// HandleGroupRatio checks for "auto_group" in the context and updates the group ratio and relayInfo.Group if present
+func HandleGroupRatio(ctx *gin.Context, relayInfo *relaycommon.RelayInfo) GroupRatioInfo {
+	groupRatioInfo := GroupRatioInfo{
+		GroupRatio:        1.0, // default ratio
+		GroupSpecialRatio: 1.0, // default user group ratio
+	}
+
+	// check auto group
+	autoGroup, exists := ctx.Get("auto_group")
+	if exists {
+		if common.DebugEnabled {
+			println(fmt.Sprintf("final group: %s", autoGroup))
+		}
+		relayInfo.Group = autoGroup.(string)
+	}
+
+	// check user group special ratio
+	userGroupRatio, ok := setting.GetGroupGroupRatio(relayInfo.UserGroup, relayInfo.Group)
+	if ok {
+		// user group special ratio
+		groupRatioInfo.GroupSpecialRatio = userGroupRatio
+		groupRatioInfo.GroupRatio = userGroupRatio
+	} else {
+		// normal group ratio
+		groupRatioInfo.GroupRatio = setting.GetGroupRatio(relayInfo.Group)
+	}
+
+	return groupRatioInfo
 }
 
 func ModelPriceHelper(c *gin.Context, info *relaycommon.RelayInfo, promptTokens int, maxTokens int) (PriceData, error) {
 	modelPrice, usePrice := operation_setting.GetModelPrice(info.OriginModelName, false)
-	groupRatio := setting.GetGroupRatio(info.Group)
-	userGroupRatio, ok := setting.GetGroupGroupRatio(info.UserGroup, info.Group)
-	if ok {
-		groupRatio = userGroupRatio
-	}
+
+	groupRatioInfo := HandleGroupRatio(c, info)
+
 	var preConsumedQuota int
 	var modelRatio float64
 	var completionRatio float64
@@ -64,18 +96,17 @@ func ModelPriceHelper(c *gin.Context, info *relaycommon.RelayInfo, promptTokens
 		cacheRatio, _ = operation_setting.GetCacheRatio(info.OriginModelName)
 		cacheCreationRatio, _ = operation_setting.GetCreateCacheRatio(info.OriginModelName)
 		imageRatio, _ = operation_setting.GetImageRatio(info.OriginModelName)
-		ratio := modelRatio * groupRatio
+		ratio := modelRatio * groupRatioInfo.GroupRatio
 		preConsumedQuota = int(float64(preConsumedTokens) * ratio)
 	} else {
-		preConsumedQuota = int(modelPrice * common.QuotaPerUnit * groupRatio)
+		preConsumedQuota = int(modelPrice * common.QuotaPerUnit * groupRatioInfo.GroupRatio)
 	}
 
 	priceData := PriceData{
 		ModelPrice:             modelPrice,
 		ModelRatio:             modelRatio,
 		CompletionRatio:        completionRatio,
-		GroupRatio:             groupRatio,
-		UserGroupRatio:         userGroupRatio,
+		GroupRatioInfo:         groupRatioInfo,
 		UsePrice:               usePrice,
 		CacheRatio:             cacheRatio,
 		ImageRatio:             imageRatio,

relay/relay-image.go

@@ -162,7 +162,7 @@ func ImageHelper(c *gin.Context) *dto.OpenAIErrorWithStatusCode {
 
 		// reset model price
 		priceData.ModelPrice *= sizeRatio * qualityRatio * float64(imageRequest.N)
-		quota = int(priceData.ModelPrice * priceData.GroupRatio * common.QuotaPerUnit)
+		quota = int(priceData.ModelPrice * priceData.GroupRatioInfo.GroupRatio * common.QuotaPerUnit)
 		userQuota, err = model.GetUserQuota(relayInfo.UserId, false)
 		if err != nil {
 			return service.OpenAIErrorWrapperLocal(err, "get_user_quota_failed", http.StatusInternalServerError)

relay/relay-text.go

@@ -90,15 +90,16 @@ func TextHelper(c *gin.Context) (openaiErr *dto.OpenAIErrorWithStatusCode) {
 
 	// get & validate textRequest 获取并验证文本请求
 	textRequest, err := getAndValidateTextRequest(c, relayInfo)
-	if textRequest.WebSearchOptions != nil {
-		c.Set("chat_completion_web_search_context_size", textRequest.WebSearchOptions.SearchContextSize)
-	}
 
 	if err != nil {
 		common.LogError(c, fmt.Sprintf("getAndValidateTextRequest failed: %s", err.Error()))
 		return service.OpenAIErrorWrapperLocal(err, "invalid_text_request", http.StatusBadRequest)
 	}
 
+	if textRequest.WebSearchOptions != nil {
+		c.Set("chat_completion_web_search_context_size", textRequest.WebSearchOptions.SearchContextSize)
+	}
+
 	if setting.ShouldCheckPromptSensitive() {
 		words, err := checkRequestSensitive(textRequest, relayInfo)
 		if err != nil {
@@ -361,9 +362,8 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 	cacheRatio := priceData.CacheRatio
 	imageRatio := priceData.ImageRatio
 	modelRatio := priceData.ModelRatio
-	groupRatio := priceData.GroupRatio
+	groupRatio := priceData.GroupRatioInfo.GroupRatio
 	modelPrice := priceData.ModelPrice
-	userGroupRatio := priceData.UserGroupRatio
 
 	// Convert values to decimal for precise calculation
 	dPromptTokens := decimal.NewFromInt(int64(promptTokens))
@@ -511,7 +511,7 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 	if extraContent != "" {
 		logContent += ", " + extraContent
 	}
-	other := service.GenerateTextOtherInfo(ctx, relayInfo, modelRatio, groupRatio, completionRatio, cacheTokens, cacheRatio, modelPrice, userGroupRatio)
+	other := service.GenerateTextOtherInfo(ctx, relayInfo, modelRatio, groupRatio, completionRatio, cacheTokens, cacheRatio, modelPrice, priceData.GroupRatioInfo.GroupSpecialRatio)
 	if imageTokens != 0 {
 		other["image"] = true
 		other["image_ratio"] = imageRatio

relay/websocket.go

@@ -6,12 +6,10 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/gorilla/websocket"
 	"net/http"
-	"one-api/common"
 	"one-api/dto"
 	relaycommon "one-api/relay/common"
+	"one-api/relay/helper"
 	"one-api/service"
-	"one-api/setting"
-	"one-api/setting/operation_setting"
 )
 
 func WssHelper(c *gin.Context, ws *websocket.Conn) (openaiErr *dto.OpenAIErrorWithStatusCode) {
@@ -39,43 +37,14 @@ func WssHelper(c *gin.Context, ws *websocket.Conn) (openaiErr *dto.OpenAIErrorWi
 			//isModelMapped = true
 		}
 	}
-	//relayInfo.UpstreamModelName = textRequest.Model
-	modelPrice, getModelPriceSuccess := operation_setting.GetModelPrice(relayInfo.UpstreamModelName, false)
-	groupRatio := setting.GetGroupRatio(relayInfo.Group)
 
-	var preConsumedQuota int
-	var ratio float64
-	var modelRatio float64
-	//err := service.SensitiveWordsCheck(textRequest)
-
-	//if constant.ShouldCheckPromptSensitive() {
-	//	err = checkRequestSensitive(textRequest, relayInfo)
-	//	if err != nil {
-	//		return service.OpenAIErrorWrapperLocal(err, "sensitive_words_detected", http.StatusBadRequest)
-	//	}
-	//}
-
-	//promptTokens, err := getWssPromptTokens(realtimeEvent, relayInfo)
-	//// count messages token error 计算promptTokens错误
-	//if err != nil {
-	//	return service.OpenAIErrorWrapper(err, "count_token_messages_failed", http.StatusInternalServerError)
-	//}
-	//
-	if !getModelPriceSuccess {
-		preConsumedTokens := common.PreConsumedQuota
-		//if realtimeEvent.Session.MaxResponseOutputTokens != 0 {
-		//	preConsumedTokens = promptTokens + int(realtimeEvent.Session.MaxResponseOutputTokens)
-		//}
-		modelRatio, _ = operation_setting.GetModelRatio(relayInfo.UpstreamModelName)
-		ratio = modelRatio * groupRatio
-		preConsumedQuota = int(float64(preConsumedTokens) * ratio)
-	} else {
-		preConsumedQuota = int(modelPrice * common.QuotaPerUnit * groupRatio)
-		relayInfo.UsePrice = true
+	priceData, err := helper.ModelPriceHelper(c, relayInfo, 0, 0)
+	if err != nil {
+		return service.OpenAIErrorWrapperLocal(err, "model_price_error", http.StatusInternalServerError)
 	}
 
 	// pre-consume quota 预消耗配额
-	preConsumedQuota, userQuota, openaiErr := preConsumeQuota(c, preConsumedQuota, relayInfo)
+	preConsumedQuota, userQuota, openaiErr := preConsumeQuota(c, priceData.ShouldPreConsumedQuota, relayInfo)
 	if openaiErr != nil {
 		return openaiErr
 	}
@@ -113,6 +82,6 @@ func WssHelper(c *gin.Context, ws *websocket.Conn) (openaiErr *dto.OpenAIErrorWi
 		return openaiErr
 	}
 	service.PostWssConsumeQuota(c, relayInfo, relayInfo.UpstreamModelName, usage.(*dto.RealtimeUsage), preConsumedQuota,
-		userQuota, modelRatio, groupRatio, modelPrice, getModelPriceSuccess, "")
+		userQuota, priceData, "")
 	return nil
 }

service/channel.go

@@ -59,6 +59,8 @@ func ShouldDisableChannel(channelType int, err *dto.OpenAIErrorWithStatusCode) b
 		return true
 	case "billing_not_active":
 		return true
+	case "pre_consume_token_quota_failed":
+		return true
 	}
 	switch err.Error.Type {
 	case "insufficient_quota":

service/error.go

@@ -29,9 +29,11 @@ func MidjourneyErrorWithStatusCodeWrapper(code int, desc string, statusCode int)
 func OpenAIErrorWrapper(err error, code string, statusCode int) *dto.OpenAIErrorWithStatusCode {
 	text := err.Error()
 	lowerText := strings.ToLower(text)
-	if strings.Contains(lowerText, "post") || strings.Contains(lowerText, "dial") || strings.Contains(lowerText, "http") {
-		common.SysLog(fmt.Sprintf("error: %s", text))
-		text = "请求上游地址失败"
+	if !strings.HasPrefix(lowerText, "get file base64 from url") && !strings.HasPrefix(lowerText, "mime type is not supported") {
+		if strings.Contains(lowerText, "post") || strings.Contains(lowerText, "dial") || strings.Contains(lowerText, "http") {
+			common.SysLog(fmt.Sprintf("error: %s", text))
+			text = "请求上游地址失败"
+		}
 	}
 	openAIError := dto.OpenAIError{
 		Message: text,
@@ -53,9 +55,11 @@ func OpenAIErrorWrapperLocal(err error, code string, statusCode int) *dto.OpenAI
 func ClaudeErrorWrapper(err error, code string, statusCode int) *dto.ClaudeErrorWithStatusCode {
 	text := err.Error()
 	lowerText := strings.ToLower(text)
-	if strings.Contains(lowerText, "post") || strings.Contains(lowerText, "dial") || strings.Contains(lowerText, "http") {
-		common.SysLog(fmt.Sprintf("error: %s", text))
-		text = "请求上游地址失败"
+	if !strings.HasPrefix(lowerText, "get file base64 from url") {
+		if strings.Contains(lowerText, "post") || strings.Contains(lowerText, "dial") || strings.Contains(lowerText, "http") {
+			common.SysLog(fmt.Sprintf("error: %s", text))
+			text = "请求上游地址失败"
+		}
 	}
 	claudeError := dto.ClaudeError{
 		Message: text,

service/file_decoder.go

@@ -4,8 +4,10 @@ import (
 	"encoding/base64"
 	"fmt"
 	"io"
+	"one-api/common"
 	"one-api/constant"
 	"one-api/dto"
+	"strings"
 )
 
 func GetFileBase64FromUrl(url string) (*dto.LocalFileData, error) {
@@ -30,9 +32,104 @@ func GetFileBase64FromUrl(url string) (*dto.LocalFileData, error) {
 	// Convert to base64
 	base64Data := base64.StdEncoding.EncodeToString(fileBytes)
 
+	mimeType := resp.Header.Get("Content-Type")
+	if len(strings.Split(mimeType, ";")) > 1 {
+		// If Content-Type has parameters, take the first part
+		mimeType = strings.Split(mimeType, ";")[0]
+	}
+	if mimeType == "application/octet-stream" {
+		if common.DebugEnabled {
+			println("MIME type is application/octet-stream, trying to guess from URL or filename")
+		}
+		// try to guess the MIME type from the url last segment
+		urlParts := strings.Split(url, "/")
+		if len(urlParts) > 0 {
+			lastSegment := urlParts[len(urlParts)-1]
+			if strings.Contains(lastSegment, ".") {
+				// Extract the file extension
+				filename := strings.Split(lastSegment, ".")
+				if len(filename) > 1 {
+					ext := strings.ToLower(filename[len(filename)-1])
+					// Guess MIME type based on file extension
+					mimeType = GetMimeTypeByExtension(ext)
+				}
+			}
+		} else {
+			// try to guess the MIME type from the file extension
+			fileName := resp.Header.Get("Content-Disposition")
+			if fileName != "" {
+				// Extract the filename from the Content-Disposition header
+				parts := strings.Split(fileName, ";")
+				for _, part := range parts {
+					if strings.HasPrefix(strings.TrimSpace(part), "filename=") {
+						fileName = strings.TrimSpace(strings.TrimPrefix(part, "filename="))
+						// Remove quotes if present
+						if len(fileName) > 2 && fileName[0] == '"' && fileName[len(fileName)-1] == '"' {
+							fileName = fileName[1 : len(fileName)-1]
+						}
+						// Guess MIME type based on file extension
+						if ext := strings.ToLower(strings.TrimPrefix(fileName, ".")); ext != "" {
+							mimeType = GetMimeTypeByExtension(ext)
+						}
+						break
+					}
+				}
+			}
+		}
+	}
+
 	return &dto.LocalFileData{
 		Base64Data: base64Data,
-		MimeType:   resp.Header.Get("Content-Type"),
+		MimeType:   mimeType,
 		Size:       int64(len(fileBytes)),
 	}, nil
 }
+
+func GetMimeTypeByExtension(ext string) string {
+	// Convert to lowercase for case-insensitive comparison
+	ext = strings.ToLower(ext)
+	switch ext {
+	// Text files
+	case "txt", "md", "markdown", "csv", "json", "xml", "html", "htm":
+		return "text/plain"
+
+	// Image files
+	case "jpg", "jpeg":
+		return "image/jpeg"
+	case "png":
+		return "image/png"
+	case "gif":
+		return "image/gif"
+
+	// Audio files
+	case "mp3":
+		return "audio/mp3"
+	case "wav":
+		return "audio/wav"
+	case "mpeg":
+		return "audio/mpeg"
+
+	// Video files
+	case "mp4":
+		return "video/mp4"
+	case "wmv":
+		return "video/wmv"
+	case "flv":
+		return "video/flv"
+	case "mov":
+		return "video/mov"
+	case "mpg":
+		return "video/mpg"
+	case "avi":
+		return "video/avi"
+	case "mpegps":
+		return "video/mpegps"
+
+	// Document files
+	case "pdf":
+		return "application/pdf"
+
+	default:
+		return "application/octet-stream" // Default for unknown types
+	}
+}

service/quota.go

@@ -3,6 +3,7 @@ package service
 import (
 	"errors"
 	"fmt"
+	"log"
 	"one-api/common"
 	constant2 "one-api/constant"
 	"one-api/dto"
@@ -94,11 +95,20 @@ func PreWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usag
 	audioInputTokens := usage.InputTokenDetails.AudioTokens
 	audioOutTokens := usage.OutputTokenDetails.AudioTokens
 	groupRatio := setting.GetGroupRatio(relayInfo.Group)
+	modelRatio, _ := operation_setting.GetModelRatio(modelName)
+
+	autoGroup, exists := ctx.Get("auto_group")
+	if exists {
+		groupRatio = setting.GetGroupRatio(autoGroup.(string))
+		log.Printf("final group ratio: %f", groupRatio)
+		relayInfo.Group = autoGroup.(string)
+	}
+
+	actualGroupRatio := groupRatio
 	userGroupRatio, ok := setting.GetGroupGroupRatio(relayInfo.UserGroup, relayInfo.Group)
 	if ok {
-		groupRatio = userGroupRatio
+		actualGroupRatio = userGroupRatio
 	}
-	modelRatio, _ := operation_setting.GetModelRatio(modelName)
 
 	quotaInfo := QuotaInfo{
 		InputDetails: TokenDetails{
@@ -112,7 +122,7 @@ func PreWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usag
 		ModelName:  modelName,
 		UsePrice:   relayInfo.UsePrice,
 		ModelRatio: modelRatio,
-		GroupRatio: groupRatio,
+		GroupRatio: actualGroupRatio,
 	}
 
 	quota := calculateAudioQuota(quotaInfo)
@@ -134,8 +144,7 @@ func PreWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usag
 }
 
 func PostWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, modelName string,
-	usage *dto.RealtimeUsage, preConsumedQuota int, userQuota int, modelRatio float64, groupRatio float64,
-	modelPrice float64, usePrice bool, extraContent string) {
+	usage *dto.RealtimeUsage, preConsumedQuota int, userQuota int, priceData helper.PriceData, extraContent string) {
 
 	useTimeSeconds := time.Now().Unix() - relayInfo.StartTime.Unix()
 	textInputTokens := usage.InputTokenDetails.TextTokens
@@ -149,11 +158,11 @@ func PostWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, mod
 	audioRatio := decimal.NewFromFloat(operation_setting.GetAudioRatio(relayInfo.OriginModelName))
 	audioCompletionRatio := decimal.NewFromFloat(operation_setting.GetAudioCompletionRatio(modelName))
 
-	actualGroupRatio := groupRatio
-	userGroupRatio, ok := setting.GetGroupGroupRatio(relayInfo.UserGroup, relayInfo.Group)
-	if ok {
-		actualGroupRatio = userGroupRatio
-	}
+	modelRatio := priceData.ModelRatio
+	groupRatio := priceData.GroupRatioInfo.GroupRatio
+	modelPrice := priceData.ModelPrice
+	usePrice := priceData.UsePrice
+
 	quotaInfo := QuotaInfo{
 		InputDetails: TokenDetails{
 			TextTokens:  textInputTokens,
@@ -166,7 +175,7 @@ func PostWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, mod
 		ModelName:  modelName,
 		UsePrice:   usePrice,
 		ModelRatio: modelRatio,
-		GroupRatio: actualGroupRatio,
+		GroupRatio: groupRatio,
 	}
 
 	quota := calculateAudioQuota(quotaInfo)
@@ -198,7 +207,7 @@ func PostWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, mod
 		logContent += ", " + extraContent
 	}
 	other := GenerateWssOtherInfo(ctx, relayInfo, usage, modelRatio, groupRatio,
-		completionRatio.InexactFloat64(), audioRatio.InexactFloat64(), audioCompletionRatio.InexactFloat64(), modelPrice, userGroupRatio)
+		completionRatio.InexactFloat64(), audioRatio.InexactFloat64(), audioCompletionRatio.InexactFloat64(), modelPrice, priceData.GroupRatioInfo.GroupSpecialRatio)
 	model.RecordConsumeLog(ctx, relayInfo.UserId, relayInfo.ChannelId, usage.InputTokens, usage.OutputTokens, logModel,
 		tokenName, quota, logContent, relayInfo.TokenId, userQuota, int(useTimeSeconds), relayInfo.IsStream, relayInfo.Group, other)
 }
@@ -214,9 +223,8 @@ func PostClaudeConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 	tokenName := ctx.GetString("token_name")
 	completionRatio := priceData.CompletionRatio
 	modelRatio := priceData.ModelRatio
-	groupRatio := priceData.GroupRatio
+	groupRatio := priceData.GroupRatioInfo.GroupRatio
 	modelPrice := priceData.ModelPrice
-	userGroupRatio := priceData.UserGroupRatio
 	cacheRatio := priceData.CacheRatio
 	cacheTokens := usage.PromptTokensDetails.CachedTokens
 
@@ -265,7 +273,7 @@ func PostClaudeConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 	}
 
 	other := GenerateClaudeOtherInfo(ctx, relayInfo, modelRatio, groupRatio, completionRatio,
-		cacheTokens, cacheRatio, cacheCreationTokens, cacheCreationRatio, modelPrice, userGroupRatio)
+		cacheTokens, cacheRatio, cacheCreationTokens, cacheCreationRatio, modelPrice, priceData.GroupRatioInfo.GroupSpecialRatio)
 	model.RecordConsumeLog(ctx, relayInfo.UserId, relayInfo.ChannelId, promptTokens, completionTokens, modelName,
 		tokenName, quota, logContent, relayInfo.TokenId, userQuota, int(useTimeSeconds), relayInfo.IsStream, relayInfo.Group, other)
 }
@@ -286,16 +294,10 @@ func PostAudioConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 	audioCompletionRatio := decimal.NewFromFloat(operation_setting.GetAudioCompletionRatio(relayInfo.OriginModelName))
 
 	modelRatio := priceData.ModelRatio
-	groupRatio := priceData.GroupRatio
+	groupRatio := priceData.GroupRatioInfo.GroupRatio
 	modelPrice := priceData.ModelPrice
 	usePrice := priceData.UsePrice
 
-	actualGroupRatio := groupRatio
-	userGroupRatio, ok := setting.GetGroupGroupRatio(relayInfo.UserGroup, relayInfo.Group)
-	if ok {
-		actualGroupRatio = userGroupRatio
-	}
-
 	quotaInfo := QuotaInfo{
 		InputDetails: TokenDetails{
 			TextTokens:  textInputTokens,
@@ -308,7 +310,7 @@ func PostAudioConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 		ModelName:  relayInfo.OriginModelName,
 		UsePrice:   usePrice,
 		ModelRatio: modelRatio,
-		GroupRatio: actualGroupRatio,
+		GroupRatio: groupRatio,
 	}
 
 	quota := calculateAudioQuota(quotaInfo)
@@ -348,7 +350,7 @@ func PostAudioConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 		logContent += ", " + extraContent
 	}
 	other := GenerateAudioOtherInfo(ctx, relayInfo, usage, modelRatio, groupRatio,
-		completionRatio.InexactFloat64(), audioRatio.InexactFloat64(), audioCompletionRatio.InexactFloat64(), modelPrice, userGroupRatio)
+		completionRatio.InexactFloat64(), audioRatio.InexactFloat64(), audioCompletionRatio.InexactFloat64(), modelPrice, priceData.GroupRatioInfo.GroupSpecialRatio)
 	model.RecordConsumeLog(ctx, relayInfo.UserId, relayInfo.ChannelId, usage.PromptTokens, usage.CompletionTokens, logModel,
 		tokenName, quota, logContent, relayInfo.TokenId, userQuota, int(useTimeSeconds), relayInfo.IsStream, relayInfo.Group, other)
 }

setting/auto_group.go

@@ -0,0 +1,31 @@
+package setting
+
+import "encoding/json"
+
+var AutoGroups = []string{
+	"default",
+}
+
+var DefaultUseAutoGroup = false
+
+func ContainsAutoGroup(group string) bool {
+	for _, autoGroup := range AutoGroups {
+		if autoGroup == group {
+			return true
+		}
+	}
+	return false
+}
+
+func UpdateAutoGroupsByJsonString(jsonString string) error {
+	AutoGroups = make([]string, 0)
+	return json.Unmarshal([]byte(jsonString), &AutoGroups)
+}
+
+func AutoGroups2JsonString() string {
+	jsonBytes, err := json.Marshal(AutoGroups)
+	if err != nil {
+		return "[]"
+	}
+	return string(jsonBytes)
+}

setting/console_setting/validation.go

@@ -7,6 +7,7 @@ import (
     "regexp"
     "strings"
     "time"
+    "sort"
 )
 
 var (
@@ -210,8 +211,23 @@ func validateFAQ(faqStr string) error {
     return nil
 }
 
+func getPublishTime(item map[string]interface{}) time.Time {
+    if v, ok := item["publishDate"]; ok {
+        if s, ok2 := v.(string); ok2 {
+            if t, err := time.Parse(time.RFC3339, s); err == nil {
+                return t
+            }
+        }
+    }
+    return time.Time{}
+}
+
 func GetAnnouncements() []map[string]interface{} {
-    return getJSONList(GetConsoleSetting().Announcements)
+    list := getJSONList(GetConsoleSetting().Announcements)
+    sort.SliceStable(list, func(i, j int) bool {
+        return getPublishTime(list[i]).After(getPublishTime(list[j]))
+    })
+    return list
 }
 
 func GetFAQ() []map[string]interface{} {

setting/operation_setting/model-ratio.go

@@ -142,6 +142,11 @@ var defaultModelRatio = map[string]float64{
 	"gemini-2.5-flash-preview-04-17":            0.075,
 	"gemini-2.5-flash-preview-04-17-thinking":   0.075,
 	"gemini-2.5-flash-preview-04-17-nothinking": 0.075,
+	"gemini-2.5-flash-preview-05-20":            0.075,
+	"gemini-2.5-flash-preview-05-20-thinking":   0.075,
+	"gemini-2.5-flash-preview-05-20-nothinking": 0.075,
+	"gemini-2.5-flash-thinking-*":               0.075, // 用于为后续所有2.5 flash thinking budget 模型设置默认倍率
+	"gemini-2.5-pro-thinking-*":                 0.625, // 用于为后续所有2.5 pro thinking budget 模型设置默认倍率
 	"text-embedding-004":                        0.001,
 	"chatglm_turbo":                             0.3572,     // ￥0.005 / 1k tokens
 	"chatglm_pro":                               0.7143,     // ￥0.01 / 1k tokens
@@ -342,10 +347,20 @@ func UpdateModelRatioByJSONString(jsonStr string) error {
 	return json.Unmarshal([]byte(jsonStr), &modelRatioMap)
 }
 
+// 处理带有思考预算的模型名称，方便统一定价
+func handleThinkingBudgetModel(name, prefix, wildcard string) string {
+	if strings.HasPrefix(name, prefix) && strings.Contains(name, "-thinking-") {
+		return wildcard
+	}
+	return name
+}
+
 func GetModelRatio(name string) (float64, bool) {
 	modelRatioMapMutex.RLock()
 	defer modelRatioMapMutex.RUnlock()
 
+	name = handleThinkingBudgetModel(name, "gemini-2.5-flash", "gemini-2.5-flash-thinking-*")
+	name = handleThinkingBudgetModel(name, "gemini-2.5-pro", "gemini-2.5-pro-thinking-*")
 	if strings.HasPrefix(name, "gpt-4-gizmo") {
 		name = "gpt-4-gizmo-*"
 	}
@@ -470,9 +485,9 @@ func getHardcodedCompletionModelRatio(name string) (float64, bool) {
 			return 4, true
 		} else if strings.HasPrefix(name, "gemini-2.0") {
 			return 4, true
-		} else if strings.HasPrefix(name, "gemini-2.5-pro-preview") {
+		} else if strings.HasPrefix(name, "gemini-2.5-pro") { // 移除preview来增加兼容性，这里假设正式版的倍率和preview一致
 			return 8, true
-		} else if strings.HasPrefix(name, "gemini-2.5-flash-preview") {
+		} else if strings.HasPrefix(name, "gemini-2.5-flash") { // 同上
 			if strings.HasSuffix(name, "-nothinking") {
 				return 4, false
 			} else {

setting/user_usable_group.go

@@ -50,3 +50,10 @@ func GroupInUserUsableGroups(groupName string) bool {
 	_, ok := userUsableGroups[groupName]
 	return ok
 }
+
+func GetUsableGroupDescription(groupName string) string {
+	if desc, ok := userUsableGroups[groupName]; ok {
+		return desc
+	}
+	return groupName
+}

web/src/components/settings/OperationSetting.js

@@ -31,6 +31,8 @@ const OperationSetting = () => {
     ModelPrice: '',
     GroupRatio: '',
     GroupGroupRatio: '',
+    AutoGroups: '',
+    DefaultUseAutoGroup: false,
     UserUsableGroups: '',
     TopUpLink: '',
     'general_setting.docs_link': '',
@@ -76,6 +78,7 @@ const OperationSetting = () => {
           item.key === 'ModelRatio' ||
           item.key === 'GroupRatio' ||
           item.key === 'GroupGroupRatio' ||
+          item.key === 'AutoGroups' ||
           item.key === 'UserUsableGroups' ||
           item.key === 'CompletionRatio' ||
           item.key === 'ModelPrice' ||
@@ -85,7 +88,8 @@ const OperationSetting = () => {
         }
         if (
           item.key.endsWith('Enabled') ||
-          ['DefaultCollapseSidebar'].includes(item.key)
+          ['DefaultCollapseSidebar'].includes(item.key) ||
+          ['DefaultUseAutoGroup'].includes(item.key)
         ) {
           newInputs[item.key] = item.value === 'true' ? true : false;
         } else {

web/src/i18n/locales/en.json

@@ -1373,6 +1373,12 @@
   "示例": "Example",
   "缺省 MaxTokens": "Default MaxTokens",
   "启用Claude思考适配（-thinking后缀）": "Enable Claude thinking adaptation (-thinking suffix)",
+  "和Claude不同，默认情况下Gemini的思考模型会自动决定要不要思考，就算不开启适配模型也可以正常使用，": "Unlike Claude, Gemini's thinking model automatically decides whether to think by default, and can be used normally even without enabling the adaptation model.",
+  "如果您需要计费，推荐设置无后缀模型价格按思考价格设置。": "If you need billing, it is recommended to set the no-suffix model price according to the thinking price.",
+  "支持使用 gemini-2.5-pro-preview-06-05-thinking-128 格式来精确传递思考预算。": "Supports using gemini-2.5-pro-preview-06-05-thinking-128 format to precisely pass thinking budget.",
+  "启用Gemini思考后缀适配": "Enable Gemini thinking suffix adaptation",
+  "适配-thinking、-thinking-预算数字和-nothinking后缀": "Adapt -thinking, -thinking-budgetNumber, and -nothinking suffixes",
+  "思考预算占比": "Thinking budget ratio",
   "Claude思考适配 BudgetTokens = MaxTokens * BudgetTokens 百分比": "Claude thinking adaptation BudgetTokens = MaxTokens * BudgetTokens percentage",
   "思考适配 BudgetTokens 百分比": "Thinking adaptation BudgetTokens percentage",
   "0.1-1之间的小数": "Decimal between 0.1 and 1",

web/src/pages/Setting/Model/SettingGeminiModel.js

@@ -173,7 +173,8 @@ export default function SettingGeminiModel(props) {
                 <Text>
                   {t(
                     "和Claude不同，默认情况下Gemini的思考模型会自动决定要不要思考，就算不开启适配模型也可以正常使用，" +
-                    "如果您需要计费，推荐设置无后缀模型价格按思考价格设置"
+                    "如果您需要计费，推荐设置无后缀模型价格按思考价格设置。" +
+                    "支持使用 gemini-2.5-pro-preview-06-05-thinking-128 格式来精确传递思考预算。"
                   )}
                 </Text>
               </Col>
@@ -183,7 +184,7 @@ export default function SettingGeminiModel(props) {
                 <Form.Switch
                   label={t('启用Gemini思考后缀适配')}
                   field={'gemini.thinking_adapter_enabled'}
-                  extraText={"适配-thinking和-nothinking后缀"}
+                  extraText={t('适配 -thinking、-thinking-预算数字 和 -nothinking 后缀')}
                   onChange={(value) =>
                     setInputs({
                       ...inputs,
@@ -205,7 +206,7 @@ export default function SettingGeminiModel(props) {
             <Row>
               <Col xs={24} sm={12} md={8} lg={8} xl={8}>
                 <Form.InputNumber
-                  label={t('请求模型带-thinking后缀的BudgetTokens数（超出24576的部分将被忽略）')}
+                  label={t('思考预算占比')}
                   field={'gemini.thinking_adapter_budget_tokens_percentage'}
                   initValue={''}
                   extraText={t('0.1-1之间的小数')}

web/src/pages/Setting/Operation/GroupRatioSettings.js

@@ -17,6 +17,8 @@ export default function GroupRatioSettings(props) {
     GroupRatio: '',
     UserUsableGroups: '',
     GroupGroupRatio: '',
+    AutoGroups: '',
+    DefaultUseAutoGroup: false,
   });
   const refForm = useRef();
   const [inputsRow, setInputsRow] = useState(inputs);
@@ -167,6 +169,59 @@ export default function GroupRatioSettings(props) {
               />
             </Col>
           </Row>
+          <Row gutter={16}>
+            <Col xs={24} sm={16}>
+              <Form.TextArea
+                label={t('自动分组auto，从第一个开始选择')}
+                placeholder={t('为一个 JSON 文本')}
+                field={'AutoGroups'}
+                autosize={{ minRows: 6, maxRows: 12 }}
+                trigger='blur'
+                stopValidateWithError
+                rules={[
+                  {
+                    validator: (rule, value) => {
+                      if (!value || value.trim() === '') {
+                        return true; // Allow empty values
+                      }
+                      
+                      // First check if it's valid JSON
+                      try {
+                        const parsed = JSON.parse(value);
+                        
+                        // Check if it's an array
+                        if (!Array.isArray(parsed)) {
+                          return false;
+                        }
+                        
+                        // Check if every element is a string
+                        return parsed.every(item => typeof item === 'string');
+                      } catch (error) {
+                        return false;
+                      }
+                    },
+                    message: t('必须是有效的 JSON 字符串数组，例如：["g1","g2"]'),
+                  },
+                ]}
+                onChange={(value) =>
+                  setInputs({ ...inputs, AutoGroups: value })
+                }
+              />
+            </Col>
+          </Row>
+          <Row gutter={16}>
+            <Col span={16}>
+              <Form.Switch
+                label={t(
+                  '创建令牌默认选择auto分组，初始令牌也将设为auto（否则留空，为用户默认分组）',
+                )}
+                field={'DefaultUseAutoGroup'}
+                onChange={(value) =>
+                  setInputs({ ...inputs, DefaultUseAutoGroup: value })
+                }
+              />
+            </Col>
+          </Row>
         </Form.Section>
       </Form>
       <Button onClick={onSubmit}>{t('保存分组倍率设置')}</Button>

web/src/pages/Token/EditToken.js

@@ -1,4 +1,4 @@
-import React, { useEffect, useState } from 'react';
+import React, { useEffect, useState, useContext } from 'react';
 import { useNavigate } from 'react-router-dom';
 import {
   API,
@@ -7,7 +7,7 @@ import {
   showSuccess,
   timestamp2string,
   renderGroupOption,
-  renderQuotaWithPrompt
+  renderQuotaWithPrompt,
 } from '../../helpers';
 import {
   AutoComplete,
@@ -37,11 +37,13 @@ import {
   IconPlusCircle,
 } from '@douyinfe/semi-icons';
 import { useTranslation } from 'react-i18next';
+import { StatusContext } from '../../context/Status';
 
 const { Text, Title } = Typography;
 
 const EditToken = (props) => {
   const { t } = useTranslation();
+  const [statusState, statusDispatch] = useContext(StatusContext);
   const [isEdit, setIsEdit] = useState(false);
   const [loading, setLoading] = useState(isEdit);
   const originInputs = {
@@ -119,7 +121,19 @@ const EditToken = (props) => {
         value: group,
         ratio: info.ratio,
       }));
+      if (statusState?.status?.default_use_auto_group) {
+        // if contain auto, add it to the first position
+        if (localGroupOptions.some((group) => group.value === 'auto')) {
+          // 排序
+          localGroupOptions.sort((a, b) => (a.value === 'auto' ? -1 : 1));
+        } else {
+          localGroupOptions.unshift({ label: t('自动选择'), value: 'auto' });
+        }
+      }
       setGroups(localGroupOptions);
+      if (statusState?.status?.default_use_auto_group) {
+        setInputs({ ...inputs, group: 'auto' });
+      }
     } else {
       showError(t(message));
     }
@@ -268,32 +282,37 @@ const EditToken = (props) => {
       placement={isEdit ? 'right' : 'left'}
       title={
         <Space>
-          {isEdit ?
-            <Tag color="blue" shape="circle">{t('更新')}</Tag> :
-            <Tag color="green" shape="circle">{t('新建')}</Tag>
-          }
-          <Title heading={4} className="m-0">
+          {isEdit ? (
+            <Tag color='blue' shape='circle'>
+              {t('更新')}
+            </Tag>
+          ) : (
+            <Tag color='green' shape='circle'>
+              {t('新建')}
+            </Tag>
+          )}
+          <Title heading={4} className='m-0'>
             {isEdit ? t('更新令牌信息') : t('创建新的令牌')}
           </Title>
         </Space>
       }
       headerStyle={{
         borderBottom: '1px solid var(--semi-color-border)',
-        padding: '24px'
+        padding: '24px',
       }}
       bodyStyle={{
         backgroundColor: 'var(--semi-color-bg-0)',
-        padding: '0'
+        padding: '0',
       }}
       visible={props.visiable}
       width={isMobile() ? '100%' : 600}
       footer={
-        <div className="flex justify-end bg-white">
+        <div className='flex justify-end bg-white'>
           <Space>
             <Button
-              theme="solid"
-              size="large"
-              className="!rounded-full"
+              theme='solid'
+              size='large'
+              className='!rounded-full'
               onClick={submit}
               icon={<IconSave />}
               loading={loading}
@@ -301,10 +320,10 @@ const EditToken = (props) => {
               {t('提交')}
             </Button>
             <Button
-              theme="light"
-              size="large"
-              className="!rounded-full"
-              type="primary"
+              theme='light'
+              size='large'
+              className='!rounded-full'
+              type='primary'
               onClick={handleCancel}
               icon={<IconClose />}
             >
@@ -317,87 +336,107 @@ const EditToken = (props) => {
       onCancel={() => handleCancel()}
     >
       <Spin spinning={loading}>
-        <div className="p-6">
-          <Card className="!rounded-2xl shadow-sm border-0 mb-6">
-            <div className="flex items-center mb-4 p-6 rounded-xl" style={{
-              background: 'linear-gradient(135deg, #1e3a8a 0%, #2563eb 50%, #3b82f6 100%)',
-              position: 'relative'
-            }}>
-              <div className="absolute inset-0 overflow-hidden">
-                <div className="absolute -top-10 -right-10 w-40 h-40 bg-white opacity-5 rounded-full"></div>
-                <div className="absolute -bottom-8 -left-8 w-24 h-24 bg-white opacity-10 rounded-full"></div>
+        <div className='p-6'>
+          <Card className='!rounded-2xl shadow-sm border-0 mb-6'>
+            <div
+              className='flex items-center mb-4 p-6 rounded-xl'
+              style={{
+                background:
+                  'linear-gradient(135deg, #1e3a8a 0%, #2563eb 50%, #3b82f6 100%)',
+                position: 'relative',
+              }}
+            >
+              <div className='absolute inset-0 overflow-hidden'>
+                <div className='absolute -top-10 -right-10 w-40 h-40 bg-white opacity-5 rounded-full'></div>
+                <div className='absolute -bottom-8 -left-8 w-24 h-24 bg-white opacity-10 rounded-full'></div>
               </div>
-              <div className="w-10 h-10 rounded-full bg-white/20 flex items-center justify-center mr-4 relative">
-                <IconPlusCircle size="large" style={{ color: '#ffffff' }} />
+              <div className='w-10 h-10 rounded-full bg-white/20 flex items-center justify-center mr-4 relative'>
+                <IconPlusCircle size='large' style={{ color: '#ffffff' }} />
               </div>
-              <div className="relative">
-                <Text style={{ color: '#ffffff' }} className="text-lg font-medium">{t('基本信息')}</Text>
-                <div style={{ color: '#ffffff' }} className="text-sm opacity-80">{t('设置令牌的基本信息')}</div>
+              <div className='relative'>
+                <Text
+                  style={{ color: '#ffffff' }}
+                  className='text-lg font-medium'
+                >
+                  {t('基本信息')}
+                </Text>
+                <div
+                  style={{ color: '#ffffff' }}
+                  className='text-sm opacity-80'
+                >
+                  {t('设置令牌的基本信息')}
+                </div>
               </div>
             </div>
 
-            <div className="space-y-4">
+            <div className='space-y-4'>
               <div>
-                <Text strong className="block mb-2">{t('名称')}</Text>
+                <Text strong className='block mb-2'>
+                  {t('名称')}
+                </Text>
                 <Input
                   placeholder={t('请输入名称')}
                   onChange={(value) => handleInputChange('name', value)}
                   value={name}
-                  autoComplete="new-password"
-                  size="large"
-                  className="!rounded-lg"
+                  autoComplete='new-password'
+                  size='large'
+                  className='!rounded-lg'
                   showClear
                   required
                 />
               </div>
 
               <div>
-                <Text strong className="block mb-2">{t('过期时间')}</Text>
-                <div className="mb-2">
+                <Text strong className='block mb-2'>
+                  {t('过期时间')}
+                </Text>
+                <div className='mb-2'>
                   <DatePicker
                     placeholder={t('请选择过期时间')}
-                    onChange={(value) => handleInputChange('expired_time', value)}
+                    onChange={(value) =>
+                      handleInputChange('expired_time', value)
+                    }
                     value={expired_time}
-                    autoComplete="new-password"
-                    type="dateTime"
-                    className="w-full !rounded-lg"
-                    size="large"
+                    autoComplete='new-password'
+                    type='dateTime'
+                    className='w-full !rounded-lg'
+                    size='large'
                     prefix={<IconCalendar />}
                   />
                 </div>
 
-                <div className="flex flex-wrap gap-2">
+                <div className='flex flex-wrap gap-2'>
                   <Button
-                    theme="light"
-                    type="primary"
+                    theme='light'
+                    type='primary'
                     onClick={() => setExpiredTime(0, 0, 0, 0)}
-                    className="!rounded-full"
+                    className='!rounded-full'
                   >
                     {t('永不过期')}
                   </Button>
                   <Button
-                    theme="light"
-                    type="tertiary"
+                    theme='light'
+                    type='tertiary'
                     onClick={() => setExpiredTime(0, 0, 1, 0)}
-                    className="!rounded-full"
+                    className='!rounded-full'
                     icon={<IconClock />}
                   >
                     {t('一小时')}
                   </Button>
                   <Button
-                    theme="light"
-                    type="tertiary"
+                    theme='light'
+                    type='tertiary'
                     onClick={() => setExpiredTime(0, 1, 0, 0)}
-                    className="!rounded-full"
+                    className='!rounded-full'
                     icon={<IconCalendar />}
                   >
                     {t('一天')}
                   </Button>
                   <Button
-                    theme="light"
-                    type="tertiary"
+                    theme='light'
+                    type='tertiary'
                     onClick={() => setExpiredTime(1, 0, 0, 0)}
-                    className="!rounded-full"
+                    className='!rounded-full'
                     icon={<IconCalendar />}
                   >
                     {t('一个月')}
@@ -407,44 +446,62 @@ const EditToken = (props) => {
             </div>
           </Card>
 
-          <Card className="!rounded-2xl shadow-sm border-0 mb-6">
-            <div className="flex items-center mb-4 p-6 rounded-xl" style={{
-              background: 'linear-gradient(135deg, #065f46 0%, #059669 50%, #10b981 100%)',
-              position: 'relative'
-            }}>
-              <div className="absolute inset-0 overflow-hidden">
-                <div className="absolute -top-10 -right-10 w-40 h-40 bg-white opacity-5 rounded-full"></div>
-                <div className="absolute -bottom-8 -left-8 w-24 h-24 bg-white opacity-10 rounded-full"></div>
+          <Card className='!rounded-2xl shadow-sm border-0 mb-6'>
+            <div
+              className='flex items-center mb-4 p-6 rounded-xl'
+              style={{
+                background:
+                  'linear-gradient(135deg, #065f46 0%, #059669 50%, #10b981 100%)',
+                position: 'relative',
+              }}
+            >
+              <div className='absolute inset-0 overflow-hidden'>
+                <div className='absolute -top-10 -right-10 w-40 h-40 bg-white opacity-5 rounded-full'></div>
+                <div className='absolute -bottom-8 -left-8 w-24 h-24 bg-white opacity-10 rounded-full'></div>
               </div>
-              <div className="w-10 h-10 rounded-full bg-white/20 flex items-center justify-center mr-4 relative">
-                <IconCreditCard size="large" style={{ color: '#ffffff' }} />
+              <div className='w-10 h-10 rounded-full bg-white/20 flex items-center justify-center mr-4 relative'>
+                <IconCreditCard size='large' style={{ color: '#ffffff' }} />
               </div>
-              <div className="relative">
-                <Text style={{ color: '#ffffff' }} className="text-lg font-medium">{t('额度设置')}</Text>
-                <div style={{ color: '#ffffff' }} className="text-sm opacity-80">{t('设置令牌可用额度和数量')}</div>
+              <div className='relative'>
+                <Text
+                  style={{ color: '#ffffff' }}
+                  className='text-lg font-medium'
+                >
+                  {t('额度设置')}
+                </Text>
+                <div
+                  style={{ color: '#ffffff' }}
+                  className='text-sm opacity-80'
+                >
+                  {t('设置令牌可用额度和数量')}
+                </div>
               </div>
             </div>
 
             <Banner
-              type="warning"
-              description={t('注意，令牌的额度仅用于限制令牌本身的最大额度使用量，实际的使用受到账户的剩余额度限制。')}
-              className="mb-4 !rounded-lg"
+              type='warning'
+              description={t(
+                '注意，令牌的额度仅用于限制令牌本身的最大额度使用量，实际的使用受到账户的剩余额度限制。',
+              )}
+              className='mb-4 !rounded-lg'
             />
 
-            <div className="space-y-4">
+            <div className='space-y-4'>
               <div>
-                <div className="flex justify-between mb-2">
+                <div className='flex justify-between mb-2'>
                   <Text strong>{t('额度')}</Text>
-                  <Text type="tertiary">{renderQuotaWithPrompt(remain_quota)}</Text>
+                  <Text type='tertiary'>
+                    {renderQuotaWithPrompt(remain_quota)}
+                  </Text>
                 </div>
                 <AutoComplete
                   placeholder={t('请输入额度')}
                   onChange={(value) => handleInputChange('remain_quota', value)}
                   value={remain_quota}
-                  autoComplete="new-password"
-                  type="number"
-                  size="large"
-                  className="w-full !rounded-lg"
+                  autoComplete='new-password'
+                  type='number'
+                  size='large'
+                  className='w-full !rounded-lg'
                   prefix={<IconCreditCard />}
                   data={[
                     { value: 500000, label: '1$' },
@@ -460,16 +517,18 @@ const EditToken = (props) => {
 
               {!isEdit && (
                 <div>
-                  <Text strong className="block mb-2">{t('新建数量')}</Text>
+                  <Text strong className='block mb-2'>
+                    {t('新建数量')}
+                  </Text>
                   <AutoComplete
                     placeholder={t('请选择或输入创建令牌的数量')}
                     onChange={(value) => handleTokenCountChange(value)}
                     onSelect={(value) => handleTokenCountChange(value)}
                     value={tokenCount.toString()}
-                    autoComplete="off"
-                    type="number"
-                    className="w-full !rounded-lg"
-                    size="large"
+                    autoComplete='off'
+                    type='number'
+                    className='w-full !rounded-lg'
+                    size='large'
                     prefix={<IconPlusCircle />}
                     data={[
                       { value: 10, label: t('10个') },
@@ -482,12 +541,12 @@ const EditToken = (props) => {
                 </div>
               )}
 
-              <div className="flex justify-end">
+              <div className='flex justify-end'>
                 <Button
-                  theme="light"
-                  type={unlimited_quota ? "danger" : "warning"}
+                  theme='light'
+                  type={unlimited_quota ? 'danger' : 'warning'}
                   onClick={setUnlimitedQuota}
-                  className="!rounded-full"
+                  className='!rounded-full'
                 >
                   {unlimited_quota ? t('取消无限额度') : t('设为无限额度')}
                 </Button>
@@ -495,92 +554,137 @@ const EditToken = (props) => {
             </div>
           </Card>
 
-          <Card className="!rounded-2xl shadow-sm border-0 mb-6">
-            <div className="flex items-center mb-4 p-6 rounded-xl" style={{
-              background: 'linear-gradient(135deg, #4c1d95 0%, #6d28d9 50%, #7c3aed 100%)',
-              position: 'relative'
-            }}>
-              <div className="absolute inset-0 overflow-hidden">
-                <div className="absolute -top-10 -right-10 w-40 h-40 bg-white opacity-5 rounded-full"></div>
-                <div className="absolute -bottom-8 -left-8 w-24 h-24 bg-white opacity-10 rounded-full"></div>
+          <Card className='!rounded-2xl shadow-sm border-0 mb-6'>
+            <div
+              className='flex items-center mb-4 p-6 rounded-xl'
+              style={{
+                background:
+                  'linear-gradient(135deg, #4c1d95 0%, #6d28d9 50%, #7c3aed 100%)',
+                position: 'relative',
+              }}
+            >
+              <div className='absolute inset-0 overflow-hidden'>
+                <div className='absolute -top-10 -right-10 w-40 h-40 bg-white opacity-5 rounded-full'></div>
+                <div className='absolute -bottom-8 -left-8 w-24 h-24 bg-white opacity-10 rounded-full'></div>
               </div>
-              <div className="w-10 h-10 rounded-full bg-white/20 flex items-center justify-center mr-4 relative">
-                <IconLink size="large" style={{ color: '#ffffff' }} />
+              <div className='w-10 h-10 rounded-full bg-white/20 flex items-center justify-center mr-4 relative'>
+                <IconLink size='large' style={{ color: '#ffffff' }} />
               </div>
-              <div className="relative">
-                <Text style={{ color: '#ffffff' }} className="text-lg font-medium">{t('访问限制')}</Text>
-                <div style={{ color: '#ffffff' }} className="text-sm opacity-80">{t('设置令牌的访问限制')}</div>
+              <div className='relative'>
+                <Text
+                  style={{ color: '#ffffff' }}
+                  className='text-lg font-medium'
+                >
+                  {t('访问限制')}
+                </Text>
+                <div
+                  style={{ color: '#ffffff' }}
+                  className='text-sm opacity-80'
+                >
+                  {t('设置令牌的访问限制')}
+                </div>
               </div>
             </div>
 
-            <div className="space-y-4">
+            <div className='space-y-4'>
               <div>
-                <Text strong className="block mb-2">{t('IP白名单')}</Text>
+                <Text strong className='block mb-2'>
+                  {t('IP白名单')}
+                </Text>
                 <TextArea
                   placeholder={t('允许的IP，一行一个，不填写则不限制')}
                   onChange={(value) => handleInputChange('allow_ips', value)}
                   value={inputs.allow_ips}
                   style={{ fontFamily: 'JetBrains Mono, Consolas' }}
-                  className="!rounded-lg"
+                  className='!rounded-lg'
                   rows={4}
                 />
-                <Text type="tertiary" className="mt-1 block text-xs">{t('请勿过度信任此功能，IP可能被伪造')}</Text>
+                <Text type='tertiary' className='mt-1 block text-xs'>
+                  {t('请勿过度信任此功能，IP可能被伪造')}
+                </Text>
               </div>
 
               <div>
-                <div className="flex items-center mb-2">
+                <div className='flex items-center mb-2'>
                   <Checkbox
                     checked={model_limits_enabled}
-                    onChange={(e) => handleInputChange('model_limits_enabled', e.target.checked)}
+                    onChange={(e) =>
+                      handleInputChange(
+                        'model_limits_enabled',
+                        e.target.checked,
+                      )
+                    }
                   >
                     <Text strong>{t('模型限制')}</Text>
                   </Checkbox>
                 </div>
                 <Select
-                  placeholder={model_limits_enabled ? t('请选择该渠道所支持的模型') : t('勾选启用模型限制后可选择')}
+                  placeholder={
+                    model_limits_enabled
+                      ? t('请选择该渠道所支持的模型')
+                      : t('勾选启用模型限制后可选择')
+                  }
                   onChange={(value) => handleInputChange('model_limits', value)}
                   value={inputs.model_limits}
                   multiple
-                  size="large"
-                  className="w-full !rounded-lg"
+                  size='large'
+                  className='w-full !rounded-lg'
                   prefix={<IconServer />}
                   optionList={models}
                   disabled={!model_limits_enabled}
                   maxTagCount={3}
                 />
-                <Text type="tertiary" className="mt-1 block text-xs">{t('非必要，不建议启用模型限制')}</Text>
+                <Text type='tertiary' className='mt-1 block text-xs'>
+                  {t('非必要，不建议启用模型限制')}
+                </Text>
               </div>
             </div>
           </Card>
 
-          <Card className="!rounded-2xl shadow-sm border-0">
-            <div className="flex items-center mb-4 p-6 rounded-xl" style={{
-              background: 'linear-gradient(135deg, #92400e 0%, #d97706 50%, #f59e0b 100%)',
-              position: 'relative'
-            }}>
-              <div className="absolute inset-0 overflow-hidden">
-                <div className="absolute -top-10 -right-10 w-40 h-40 bg-white opacity-5 rounded-full"></div>
-                <div className="absolute -bottom-8 -left-8 w-24 h-24 bg-white opacity-10 rounded-full"></div>
+          <Card className='!rounded-2xl shadow-sm border-0'>
+            <div
+              className='flex items-center mb-4 p-6 rounded-xl'
+              style={{
+                background:
+                  'linear-gradient(135deg, #92400e 0%, #d97706 50%, #f59e0b 100%)',
+                position: 'relative',
+              }}
+            >
+              <div className='absolute inset-0 overflow-hidden'>
+                <div className='absolute -top-10 -right-10 w-40 h-40 bg-white opacity-5 rounded-full'></div>
+                <div className='absolute -bottom-8 -left-8 w-24 h-24 bg-white opacity-10 rounded-full'></div>
               </div>
-              <div className="w-10 h-10 rounded-full bg-white/20 flex items-center justify-center mr-4 relative">
-                <IconUserGroup size="large" style={{ color: '#ffffff' }} />
+              <div className='w-10 h-10 rounded-full bg-white/20 flex items-center justify-center mr-4 relative'>
+                <IconUserGroup size='large' style={{ color: '#ffffff' }} />
               </div>
-              <div className="relative">
-                <Text style={{ color: '#ffffff' }} className="text-lg font-medium">{t('分组信息')}</Text>
-                <div style={{ color: '#ffffff' }} className="text-sm opacity-80">{t('设置令牌的分组')}</div>
+              <div className='relative'>
+                <Text
+                  style={{ color: '#ffffff' }}
+                  className='text-lg font-medium'
+                >
+                  {t('分组信息')}
+                </Text>
+                <div
+                  style={{ color: '#ffffff' }}
+                  className='text-sm opacity-80'
+                >
+                  {t('设置令牌的分组')}
+                </div>
               </div>
             </div>
 
             <div>
-              <Text strong className="block mb-2">{t('令牌分组')}</Text>
+              <Text strong className='block mb-2'>
+                {t('令牌分组')}
+              </Text>
               {groups.length > 0 ? (
                 <Select
                   placeholder={t('令牌分组，默认为用户的分组')}
                   onChange={(value) => handleInputChange('group', value)}
                   renderOptionItem={renderGroupOption}
                   value={inputs.group}
-                  size="large"
-                  className="w-full !rounded-lg"
+                  size='large'
+                  className='w-full !rounded-lg'
                   prefix={<IconUserGroup />}
                   optionList={groups}
                 />
@@ -588,8 +692,8 @@ const EditToken = (props) => {
                 <Select
                   placeholder={t('管理员未设置用户可选分组')}
                   disabled={true}
-                  size="large"
-                  className="w-full !rounded-lg"
+                  size='large'
+                  className='w-full !rounded-lg'
                   prefix={<IconUserGroup />}
                 />
               )}