diff --git a/model/domain/schema_mapping.go b/model/domain/schema_mapping.go
index e787af4..4f32fc5 100644
--- a/model/domain/schema_mapping.go
+++ b/model/domain/schema_mapping.go
@@ -44,3 +44,9 @@ type VideoFields struct {
 	NegativePrompt         string `json:"negative_prompt" dc:"反向提示词，描述不希望出现的内容"`
 	CfgScale               string `json:"cfg_scale" dc:"CFG 引导比例，控制对 prompt 的遵从程度"`
 }
+
+// VideoFieldsRes 视频模型业务字段映射
+// 适用于 视频模型(600) 及其子类型
+type VideoFieldsRes struct {
+	Duration int64 `json:"duration" dc:"视频时长（秒）"`
+}
diff --git a/model/dto/model_call_dto.go b/model/dto/model_call_dto.go
index f33a2b8..55cb970 100644
--- a/model/dto/model_call_dto.go
+++ b/model/dto/model_call_dto.go
@@ -66,6 +66,7 @@ type ModelMsg struct {
 	CompletionTokens int64          `json:"completionTokens" dc:"输出token"`
 	Content          map[string]any `json:"content" dc:"内容"`
 	Cost             float64        `json:"cost" dc:"费用（元）"`
+	Duration         int64          `json:"duration" dc:"时长（秒）"`
 	ErrorMsg         string         `json:"errorMsg" dc:"错误消息"`
 }
 
diff --git a/service/model_call_service.go b/service/model_call_service.go
index de45342..c4a9d02 100644
--- a/service/model_call_service.go
+++ b/service/model_call_service.go
@@ -34,7 +34,7 @@ func (s *modelCallService) ModelCall(ctx context.Context, req *dto.ModelCallReq)
 		return nil, fmt.Errorf("模型不存在")
 	}
 	// 引用行 → 解析为系统模型配置+本人 apiKey；系统模型已删除等解析失败 → 阻塞调用
-	modelInfo, err = resolveModelConfig(ctx, modelInfo)
+	modelInfo, err = modelUtils.ResolveModelConfig(ctx, modelInfo)
 	if err != nil {
 		return nil, err
 	}
@@ -47,9 +47,11 @@ func (s *modelCallService) ModelCall(ctx context.Context, req *dto.ModelCallReq)
 	if err != nil {
 		return
 	}
-	// 调用前检查模型计价配置（shop-user-trade）：未配置/未启用 → 阻塞调用（subject=解析后的系统模型 id）
-	if err = modelBillable(ctx, modelInfo.Id); err != nil {
-		return nil, err
+	if !g.IsEmpty(modelInfo.RefSystemModelId) {
+		// 调用前检查模型计价配置（shop-user-trade）：未配置/未启用 → 阻塞调用（subject=解析后的系统模型 id）
+		if err = modelBillable(ctx, modelInfo.Id); err != nil {
+			return nil, err
+		}
 	}
 	err = queue(ctx, modelInfo.ModelName, userInfo.TenantId, gconv.Int64(modelInfo.MaxConcurrency), func(ctx context.Context) (err error) {
 		if *modelInfo.ResponseType == *model.ResponseTypeSync.Code() || *modelInfo.ResponseType == *model.ResponseTypeStream.Code() {
@@ -113,7 +115,7 @@ func (s *modelCallService) ModelCallStream(ctx context.Context, w http.ResponseW
 		return fmt.Errorf("模型不存在")
 	}
 	// 引用行 → 解析为系统模型配置+本人 apiKey；系统模型已删除等解析失败 → 阻塞调用
-	modelInfo, err = resolveModelConfig(ctx, modelInfo)
+	modelInfo, err = modelUtils.ResolveModelConfig(ctx, modelInfo)
 	if err != nil {
 		return err
 	}
@@ -126,9 +128,11 @@ func (s *modelCallService) ModelCallStream(ctx context.Context, w http.ResponseW
 		if err != nil {
 			return err
 		}
-		// 调用前检查模型计价配置（shop-user-trade）：未配置/未启用 → 阻塞调用（subject=解析后的系统模型 id）
-		if err = modelBillable(ctx, modelInfo.Id); err != nil {
-			return err
+		if !g.IsEmpty(modelInfo.RefSystemModelId) {
+			// 调用前检查模型计价配置（shop-user-trade）：未配置/未启用 → 阻塞调用（subject=解析后的系统模型 id）
+			if err = modelBillable(ctx, modelInfo.Id); err != nil {
+				return err
+			}
 		}
 		err = queue(ctx, modelInfo.ModelName, userInfo.TenantId, gconv.Int64(modelInfo.MaxConcurrency), func(ctx context.Context) (err error) {
 			var newRequestParams map[string]any
@@ -201,7 +205,7 @@ func (s *modelCallService) saveModelRequestParams(ctx context.Context, now time.
 			MsgTopic:            req.MsgTopic,
 			RequestPath:         uploadNewReq.FileURL,
 			OriginalRequestPath: uploadOriginalReq.FileURL,
-			MediaType:           DetectMediaType(modelInfo.RequestBusinessFieldMapping, out),
+			MediaType:           modelUtils.DetectMediaType(modelInfo.RequestBusinessFieldMapping, out),
 		})
 		if err != nil {
 			return 0, nil, fmt.Errorf("保存模型请求信息失败: %v", err)
diff --git a/service/model_http_service.go b/service/model_http_service.go
index 5afd78e..e5093bb 100644
--- a/service/model_http_service.go
+++ b/service/model_http_service.go
@@ -7,13 +7,11 @@ import (
 	commonHttp "gitea.redpowerfuture.com/red-future/common/http"
 	"gitea.redpowerfuture.com/red-future/common/oss"
 	"gitea.redpowerfuture.com/red-future/common/utils"
-	"github.com/gogf/gf/v2/frame/g"
-	"github.com/gogf/gf/v2/util/gconv"
 )
 
 // IsSuperAdmin 调用admin-go服务检查是否是管理员
 func IsSuperAdmin(ctx context.Context) (res bool, err error) {
-	headers := setCtxHeader(ctx)
+	headers := utils.HeadersFromCtx(ctx, utils.HeadersOptions{ResolveToken: true})
 	var r = make(map[string]bool)
 	if err = commonHttp.Get(ctx, "admin-go/api/v1/system/user/checkIsSuperAdmin", headers, &r); err != nil {
 		return false, err
@@ -35,29 +33,3 @@ func Upload(ctx context.Context, req *dto.UploadFileBytesReq) (*dto.UploadFileBy
 		FileAddressPrefix: res.FileAddressPrefix,
 	}, nil
 }
-
-// setCtxHeader 构造调用方请求头透传（X-User-Info 三态注入）：
-//  1. 透传 HTTP 请求头（含 Authorization/X-User-Info）
-//  2. ctx 无请求头时，用任务体注入的 user（异步任务 Creator/TenantId）生成 X-User-Info
-//  3. 仍为空时，解析调用方 token 得到用户生成 X-User-Info（直连场景归属校验）
-func setCtxHeader(ctx context.Context) map[string]string {
-	headers := make(map[string]string)
-	if r := g.RequestFromCtx(ctx); r != nil {
-		for k, v := range r.Request.Header {
-			if len(v) > 0 {
-				headers[k] = v[0]
-			}
-		}
-	}
-	if headers["X-User-Info"] == "" {
-		if user := ctx.Value("user"); !g.IsNil(user) {
-			headers["X-User-Info"] = gconv.String(user)
-		}
-	}
-	if headers["X-User-Info"] == "" {
-		if user, err := utils.GetUserInfo(ctx); err == nil && user != nil {
-			headers["X-User-Info"] = gconv.String(user)
-		}
-	}
-	return headers
-}
diff --git a/service/model_manage_service.go b/service/model_manage_service.go
index 7a245bc..06469f7 100644
--- a/service/model_manage_service.go
+++ b/service/model_manage_service.go
@@ -8,6 +8,7 @@ import (
 	"model-gateway/dao"
 	"model-gateway/model/dto"
 	"model-gateway/model/entity"
+	modelUtils "model-gateway/service/utils"
 
 	"gitea.redpowerfuture.com/red-future/common/beans"
 	"gitea.redpowerfuture.com/red-future/common/db/gfdb"
@@ -268,7 +269,7 @@ func (s *modelManageService) Get(ctx context.Context, req *dto.GetModelManageReq
 			return nil, e
 		}
 		if sys != nil {
-			get = mergeReferenceConfigForQuery(get, sys)
+			get = modelUtils.MergeReferenceConfigForQuery(get, sys)
 		}
 	}
 	// 系统模型 apiKey 对非创建者脱敏（引用行/自有行仅本人可见）
@@ -324,6 +325,25 @@ func (s *modelManageService) List(ctx context.Context, req *dto.ListModelManageR
 	if err != nil {
 		return
 	}
+	// 引用行 → 合入系统模型配置返回（与 Get 展示一致；保留引用行自身 Id/SystemModel 供更新/脱敏判断）
+	sysCache := make(map[int64]*entity.ModelManage)
+	for _, row := range list {
+		if row.RefSystemModelId <= 0 {
+			continue
+		}
+		sys, ok := sysCache[row.RefSystemModelId]
+		if !ok {
+			sys, err = dao.ModelManage.GetNotTenantId(ctx, &dto.GetModelManageReq{Id: row.RefSystemModelId})
+			if err != nil {
+				return
+			}
+			if sys == nil {
+				continue
+			}
+			sysCache[row.RefSystemModelId] = sys
+		}
+		*row = *modelUtils.MergeReferenceConfigForQuery(row, sys)
+	}
 	// 系统模型 apiKey 对非创建者脱敏（引用行/自有行仅本人可见）
 	for _, row := range list {
 		if row.SystemModel != nil && *row.SystemModel && row.Creator != user.UserName {
diff --git a/service/model_task_end_service.go b/service/model_task_end_service.go
index 554387f..e4cbb9b 100644
--- a/service/model_task_end_service.go
+++ b/service/model_task_end_service.go
@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"model-gateway/consts/public"
 	"model-gateway/dao"
+	"model-gateway/model/domain"
 	"model-gateway/model/dto"
 	"model-gateway/model/entity"
 	"model-gateway/service/httpclient"
@@ -199,7 +200,7 @@ func (s *modelTaskEndService) processClaimedTask(asyncCtx context.Context, item
 	}
 
 	// 引用行 → 解析为系统模型配置+本人 apiKey（轮询/计价均用系统模型）
-	modelInfo, err = resolveModelConfig(asyncCtx, modelInfo)
+	modelInfo, err = modelUtils.ResolveModelConfig(asyncCtx, modelInfo)
 	if err != nil {
 		g.Log().Errorf(asyncCtx, "模型配置解析失败: modelId=%d err=%v", item.ModelId, err)
 		docMsg.ErrorMsg = fmt.Sprintf("模型配置解析失败: %v", err)
@@ -281,6 +282,20 @@ LOOP:
 		}
 		docMsg.Content = content
 
+		// 解析 ResponseBusinessFieldMapping 字段
+		businessField := make(map[string]any, len(modelInfo.ResponseBusinessFieldMapping))
+		for key, value := range modelInfo.ResponseBusinessFieldMapping {
+			businessField[key] = modelUtils.GetByPathValue(respObj, modelUtils.CleanFieldPath(value))
+		}
+		businessFieldRes := new(domain.VideoFieldsRes)
+		err = gconv.Struct(businessField, businessFieldRes)
+		if err != nil {
+			docMsg.ErrorMsg = fmt.Sprintf("解析 ResponseBusinessFieldMapping 字段失败: %v", err)
+			finalize()
+			return nil
+		}
+		docMsg.Duration = businessFieldRes.Duration
+
 		// 解析Token
 		totalTokPath := modelUtils.CleanFieldPath(modelInfo.TokenMapping.TotalTokens)
 		promptTokPath := modelUtils.CleanFieldPath(modelInfo.TokenMapping.PromptTokens)
@@ -290,10 +305,6 @@ LOOP:
 		docMsg.PromptTokens = gconv.Int64(modelUtils.GetByPathValue(respObj, promptTokPath))
 		docMsg.CompletionTokens = gconv.Int64(modelUtils.GetByPathValue(respObj, compTokPath))
 
-		// 调 shop-user-trade 按用量算费（媒体类型取任务创建时的快照；subject=解析后的系统模型 id）
-		docMsg.Cost = calcModelCost(asyncCtx, modelInfo.Id,
-			buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, item.MediaType, time.Since(startTime).Seconds()))
-
 		// 判断任务状态，轮询等待
 		statusPath := modelUtils.CleanFieldPath(modelInfo.AsyncTaskMapping.TaskStatus)
 		status := gconv.String(modelUtils.GetByPathValue(respObj, statusPath))
@@ -301,6 +312,10 @@ LOOP:
 			time.Sleep(10 * time.Second)
 			goto LOOP
 		}
+
+		// 调 shop-user-trade 按用量算费（媒体类型取任务创建时的快照；subject=解析后的系统模型 id）
+		docMsg.Cost = calcModelCost(asyncCtx, modelInfo.Id,
+			buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, item.MediaType, docMsg.Duration))
 	}
 	// 成功或已识别出错误的终态统一落库+发布（内容组装完成/错误消息已写入 docMsg）
 	finalize()
diff --git a/service/parse_error_test.go b/service/parse_error_test.go
deleted file mode 100644
index ee33c48..0000000
--- a/service/parse_error_test.go
+++ /dev/null
@@ -1,146 +0,0 @@
-package service
-
-import "testing"
-
-// flatMapping 扁平形态：code 配置 defaultValue=20000000 为成功码
-func flatMapping() map[string]any {
-	return map[string]any{
-		"code": map[string]any{
-			"type": "number", "value": 0, "label": "", "fieldType": "number", "isForm": false, "required": false,
-			"defaultValue": float64(20000000),
-		},
-		"message": map[string]any{
-			"type": "string", "value": "", "label": "", "fieldType": "string", "isForm": false, "required": false,
-		},
-	}
-}
-
-func TestParseModelErrorFlatDefaultValue(t *testing.T) {
-	mapping := flatMapping()
-	// 成功：code == defaultValue
-	code, msg, err := parseModelError([]byte(`{"code":20000000,"message":"ok"}`), mapping)
-	if err != nil || code != "" || msg != "" {
-		t.Fatalf("success(default): code=%q msg=%q err=%v", code, msg, err)
-	}
-	// 成功：code 数字零值（视为空）
-	code, msg, err = parseModelError([]byte(`{"code":0}`), mapping)
-	if err != nil || code != "" || msg != "" {
-		t.Fatalf("success(zero): code=%q msg=%q err=%v", code, msg, err)
-	}
-	// 错误：code != defaultValue
-	code, msg, err = parseModelError([]byte(`{"code":1234,"message":"boom"}`), mapping)
-	if err != nil || code != "1234" || msg != "boom" {
-		t.Fatalf("error: code=%q msg=%q err=%v", code, msg, err)
-	}
-}
-
-func TestParseModelErrorNested(t *testing.T) {
-	mapping := map[string]any{
-		"error": map[string]any{
-			"type": "object",
-			"attrs": map[string]any{
-				"code":    map[string]any{"type": "string", "label": "", "value": "", "fieldType": "string", "isForm": false, "required": false},
-				"message": map[string]any{"type": "string", "label": "", "value": "", "fieldType": "string", "isForm": false, "required": false},
-			},
-			"label": "", "isForm": false, "required": false, "fieldType": "string",
-		},
-	}
-	// 错误响应
-	code, msg, err := parseModelError([]byte(`{"error":{"code":"rate_limit_exceeded","message":"Too fast"}}`), mapping)
-	if err != nil || code != "rate_limit_exceeded" || msg != "Too fast" {
-		t.Fatalf("error: code=%q msg=%q err=%v", code, msg, err)
-	}
-	// 成功响应（无 error 字段 → code 提取为空）
-	code, msg, err = parseModelError([]byte(`{"id":"x","choices":[]}`), mapping)
-	if err != nil || code != "" || msg != "" {
-		t.Fatalf("success: code=%q msg=%q err=%v", code, msg, err)
-	}
-}
-
-func TestParseModelErrorArray(t *testing.T) {
-	mapping := map[string]any{
-		"errors": map[string]any{
-			"type":  "array",
-			"attrs": []any{map[string]any{
-				"code":    map[string]any{"type": "number"},
-				"message": map[string]any{"type": "string"},
-			}},
-		},
-	}
-	code, msg, err := parseModelError([]byte(`{"errors":[{"code":1001,"message":"err-a"}]}`), mapping)
-	if err != nil || code != "1001" || msg != "err-a" {
-		t.Fatalf("array error: code=%q msg=%q err=%v", code, msg, err)
-	}
-	code, msg, err = parseModelError([]byte(`{"errors":[]}`), mapping)
-	if err != nil || code != "" || msg != "" {
-		t.Fatalf("array success: code=%q msg=%q err=%v", code, msg, err)
-	}
-}
-
-func TestParseModelErrorPlainStringMapping(t *testing.T) {
-	// 纯字符串路径形态：字段值直接是字段路径
-	mapping := map[string]any{"code": "error.code", "message": "error.message"}
-	code, msg, err := parseModelError([]byte(`{"error":{"code":100,"message":"nope"}}`), mapping)
-	if err != nil || code != "100" || msg != "nope" {
-		t.Fatalf("plain mapping: code=%q msg=%q err=%v", code, msg, err)
-	}
-	code, msg, err = parseModelError([]byte(`{"data":"ok"}`), mapping)
-	if err != nil || code != "" || msg != "" {
-		t.Fatalf("plain success: code=%q msg=%q err=%v", code, msg, err)
-	}
-}
-
-func TestParseModelErrorNoMapping(t *testing.T) {
-	// 未配置 mapping：不识别错误，一律成功（纯配置驱动）
-	code, msg, err := parseModelError([]byte(`{"error":{"code":"x"}}`), nil)
-	if err != nil || code != "" || msg != "" {
-		t.Fatalf("no mapping: code=%q msg=%q err=%v", code, msg, err)
-	}
-}
-
-func TestParseModelErrorBadJSON(t *testing.T) {
-	mapping := map[string]any{"code": map[string]any{"type": "string"}}
-	if _, _, err := parseModelError([]byte(`{invalid`), mapping); err == nil {
-		t.Fatalf("bad json should err")
-	}
-}
-
-func TestParseModelErrorOnlyMessage(t *testing.T) {
-	// 只配置 message：以 message 是否为空判定错误
-	mapping := map[string]any{"message": map[string]any{"type": "string"}}
-	code, msg, err := parseModelError([]byte(`{"message":"something went wrong"}`), mapping)
-	if err != nil || code != "" || msg != "something went wrong" {
-		t.Fatalf("only-message error: code=%q msg=%q err=%v", code, msg, err)
-	}
-	code, msg, err = parseModelError([]byte(`{"data":"ok"}`), mapping)
-	if err != nil || code != "" || msg != "" {
-		t.Fatalf("only-message success: code=%q msg=%q err=%v", code, msg, err)
-	}
-}
-
-func TestCollectErrorMapping(t *testing.T) {
-	// 扁平 + defaultValue
-	codePath, msgPath, hasDef, def := collectErrorMapping(flatMapping())
-	if codePath != "code" || msgPath != "message" || !hasDef || def != float64(20000000) {
-		t.Fatalf("flat: code=%q msg=%q hasDef=%v def=%v", codePath, msgPath, hasDef, def)
-	}
-	// 嵌套
-	codePath, msgPath, hasDef, def = collectErrorMapping(map[string]any{
-		"error": map[string]any{"type": "object", "attrs": map[string]any{
-			"code":    map[string]any{"type": "string"},
-			"message": map[string]any{"type": "string"},
-		}},
-	})
-	if codePath != "error.code" || msgPath != "error.message" || hasDef {
-		t.Fatalf("nested: code=%q msg=%q hasDef=%v", codePath, msgPath, hasDef)
-	}
-	// 数组 → [*]
-	codePath, _, _, _ = collectErrorMapping(map[string]any{
-		"errors": map[string]any{"type": "array", "attrs": []any{
-			map[string]any{"code": map[string]any{"type": "number"}},
-		}},
-	})
-	if codePath != "errors[*].code" {
-		t.Fatalf("array: code=%q", codePath)
-	}
-}
diff --git a/service/pricing_client.go b/service/pricing_client.go
index ffd6663..12c0f60 100644
--- a/service/pricing_client.go
+++ b/service/pricing_client.go
@@ -5,14 +5,25 @@ import (
 	"fmt"
 
 	commonHttp "gitea.redpowerfuture.com/red-future/common/http"
+	"gitea.redpowerfuture.com/red-future/common/utils"
 	"github.com/gogf/gf/v2/frame/g"
 )
 
 // ====================== shop-user-trade 计价对接（独立 module，本地 JSON 对齐） ======================
 
-// shopPricingConfig shop-user-trade config/get 响应（仅取启用标记；规则在 shop-user-trade 侧，model-gateway 不消费）
+// shopPricingConfig shop-user-trade config/get 响应（仅取启用标记与最低余额门限；费率规则在 shop-user-trade 侧，model-gateway 不消费）
 type shopPricingConfig struct {
-	Enabled int `json:"enabled"`
+	Enabled    int     `json:"enabled"`
+	MinBalance float64 `json:"minBalance" dc:"门禁：调用前可用余额须>=该值（元），0=不校验"`
+}
+
+// shopWalletAccount shop-user-trade wallet/account/get 响应
+type shopWalletAccount struct {
+	ID       int64   `json:"id"`
+	UserID   int64   `json:"userId"`
+	Balance  float64 `json:"balance" dc:"可用余额（元），负值=欠费"`
+	Currency string  `json:"currency"`
+	Status   int     `json:"status" dc:"状态：1启用 0禁用 -1冻结"`
 }
 
 // shopCalcFeeRes shop-user-trade /calc 响应
@@ -37,11 +48,17 @@ func mgMediaTypeToShop(mg string) string {
 	}
 }
 
+// walletURL 组装 shop-user-trade 钱包接口地址（accountController → account/controller，与 pricing 同 RouteRegister 推导规则）
+func walletURL(sub string) string {
+	return "shop-user-trade/account/controller/" + sub
+}
+
 // modelBillable 调用前门禁：模型须在 shop-user-trade 已配置且启用计价，否则阻塞调用。
 // 替换原 CheckTenantBalance（admin-go 租户余额门禁）；未配置→config/get 返回错误，未启用→enabled!=1。
+// minBalance>0 时追加最低余额门禁：钱包须存在且可用余额 >= 门限（与 shop-user-trade open_order 同语义，0=不校验）。
 func modelBillable(ctx context.Context, modelId int64) error {
 	var cfg shopPricingConfig
-	err := commonHttp.Get(ctx, pricingURL("config/get"), setCtxHeader(ctx), &cfg,
+	err := commonHttp.Get(ctx, pricingURL("config/get"), utils.HeadersFromCtx(ctx, utils.HeadersOptions{ResolveToken: true}), &cfg,
 		"subjectType", "model", "subjectId", fmt.Sprintf("%d", modelId))
 	if err != nil {
 		return fmt.Errorf("模型未配置计价，无法调用: %w", err)
@@ -49,11 +66,29 @@ func modelBillable(ctx context.Context, modelId int64) error {
 	if cfg.Enabled != 1 {
 		return fmt.Errorf("模型未启用计价，无法调用")
 	}
+	if cfg.MinBalance <= 0 {
+		return nil
+	}
+	user, e := utils.GetUserInfo(ctx)
+	if e != nil || user == nil || user.Id == 0 {
+		return fmt.Errorf("取不到用户，无法校验最低余额")
+	}
+	var acc shopWalletAccount
+	if e = commonHttp.Get(ctx, walletURL("get"), utils.HeadersFromCtx(ctx, utils.HeadersOptions{ResolveToken: true}), &acc,
+		"userId", fmt.Sprintf("%d", user.Id)); e != nil {
+		return fmt.Errorf("获取钱包失败，无法校验最低余额: %w", e)
+	}
+	if acc.Status != 1 {
+		return fmt.Errorf("钱包不可用，无法调用")
+	}
+	if acc.Balance < cfg.MinBalance {
+		return fmt.Errorf("余额不足：可用余额须不低于 %.2f 元才能发起", cfg.MinBalance)
+	}
 	return nil
 }
 
 // buildModelUsage 组装算费用量 JSON 对象（ChargeUsage 形状）。mediaType 为 model-gateway 词汇（DetectMediaType/异步快照）。
-func buildModelUsage(prompt, completion, cached int64, mediaType string, durationSec float64) map[string]any {
+func buildModelUsage(prompt, completion, cached int64, mediaType string, durationSec int64) map[string]any {
 	if durationSec < 0 {
 		durationSec = 0
 	}
@@ -70,7 +105,7 @@ func buildModelUsage(prompt, completion, cached int64, mediaType string, duratio
 // 调用前门禁已保证配置存在；此处失败（配置中途删除/网络抖动）→ 记日志返回 0，不拖垮已完成的模型调用。
 func calcModelCost(ctx context.Context, modelId int64, usage map[string]any) float64 {
 	var res shopCalcFeeRes
-	err := commonHttp.Post(ctx, pricingURL("calc"), setCtxHeader(ctx), &res, &struct {
+	err := commonHttp.Post(ctx, pricingURL("calc"), utils.HeadersFromCtx(ctx, utils.HeadersOptions{ResolveToken: true}), &res, &struct {
 		SubjectType string         `json:"subjectType"`
 		SubjectID   string         `json:"subjectId"`
 		Usage       map[string]any `json:"usage"`
diff --git a/service/session_stream.go b/service/session_stream.go
index 750b16f..3be7528 100644
--- a/service/session_stream.go
+++ b/service/session_stream.go
@@ -127,9 +127,9 @@ LOOP:
 	}
 	updateModelSessionReq.DurationSeconds = int64(time.Since(startTime).Seconds())
 	// 调 shop-user-trade 按用量算费（不本地换算；调用前门禁已保证配置存在，失败→0 不阻塞）
-	mediaType := DetectMediaType(modelInfo.RequestBusinessFieldMapping, newRequestParams)
+	mediaType := modelUtils.DetectMediaType(modelInfo.RequestBusinessFieldMapping, newRequestParams)
 	docMsg.Cost = calcModelCost(ctx, modelInfo.Id,
-		buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, mediaType, time.Since(startTime).Seconds()))
+		buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, mediaType, 0))
 	updateModelSessionReq.TotalCost = docMsg.Cost
 	if _, updateErr := dao.ModelSession.Update(ctx, &updateModelSessionReq); updateErr != nil {
 		g.Log().Errorf(ctx, "更新流式会话信息失败: %v", updateErr)
@@ -238,9 +238,9 @@ func (s *modelSessionService) CreateSessionStream(ctx context.Context, w http.Re
 	})
 
 	// 流结束：调 shop-user-trade 按用量算费（不本地换算；调用前门禁已保证配置存在，失败→0 不阻塞）
-	mediaType := DetectMediaType(modelInfo.RequestBusinessFieldMapping, newRequestParams)
+	mediaType := modelUtils.DetectMediaType(modelInfo.RequestBusinessFieldMapping, newRequestParams)
 	docMsg.Cost = calcModelCost(ctx, modelInfo.Id,
-		buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, mediaType, time.Since(startTime).Seconds()))
+		buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, mediaType, 0))
 
 	// 流末 done 事件：携带该步最终 token 与费用。工具调用时附带完整 tool_calls，
 	// 纯文本流同样补发，使调用方拿到最终费用与 token；不识别 type=done 的消费方忽略该事件。
diff --git a/service/session_sync.go b/service/session_sync.go
index 4769821..eabb093 100644
--- a/service/session_sync.go
+++ b/service/session_sync.go
@@ -121,9 +121,9 @@ LOOP:
 	}
 	updateModelSessionReq.DurationSeconds = int64(time.Since(startTime).Seconds())
 	// 9.5) 调 shop-user-trade 按用量算费（不本地换算；调用前门禁已保证配置存在，失败→0 不阻塞）
-	mediaType := DetectMediaType(modelInfo.RequestBusinessFieldMapping, newRequestParams)
+	mediaType := modelUtils.DetectMediaType(modelInfo.RequestBusinessFieldMapping, newRequestParams)
 	docMsg.Cost = calcModelCost(ctx, modelInfo.Id,
-		buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, mediaType, time.Since(startTime).Seconds()))
+		buildModelUsage(docMsg.PromptTokens, docMsg.CompletionTokens, 0, mediaType, 0))
 	updateModelSessionReq.TotalCost = docMsg.Cost
 	// 10) 更新模型会话信息
 	_, err = dao.ModelSession.Update(ctx, &updateModelSessionReq)
diff --git a/service/price.go b/service/utils/media_type.go
similarity index 82%
rename from service/price.go
rename to service/utils/media_type.go
index 8d75bc4..57ab5bb 100644
--- a/service/price.go
+++ b/service/utils/media_type.go
@@ -1,8 +1,4 @@
-package service
-
-import (
-	modelUtils "model-gateway/service/utils"
-)
+package utils
 
 // DetectMediaType 按模型业务字段映射从请求体推导输入媒体类型（替代硬编码的 media.type 路径）：
 //   - reference_audio 映射路径在请求体中有值 → "audio"
@@ -11,7 +7,7 @@ import (
 //
 // 判定完全由模型配置（RequestBusinessFieldMapping，业务字段名见 ChatFieldsReq/VideoFields）驱动，
 // 无请求结构硬编码；映射路径值即 GetByPathAll 路径（如 input.media?type=audio&url=#）。
-// 媒体类型仅供 shop-user-trade 算费用量（见 pricing_client.go mgMediaTypeToShop）。
+// 媒体类型仅供 shop-user-trade 算费用量（见 service/pricing_client.go mgMediaTypeToShop）。
 func DetectMediaType(reqBizMapping map[string]string, reqParams map[string]any) string {
 	if hasMediaValue(reqBizMapping, reqParams, "reference_audio") {
 		return "audio"
@@ -28,5 +24,5 @@ func hasMediaValue(reqBizMapping map[string]string, reqParams map[string]any, bi
 	if path == "" {
 		return false
 	}
-	return len(modelUtils.GetByPathAll(reqParams, path)) > 0
+	return len(GetByPathAll(reqParams, path)) > 0
 }
diff --git a/service/model_resolve.go b/service/utils/model_resolve.go
similarity index 92%
rename from service/model_resolve.go
rename to service/utils/model_resolve.go
index 1a6b2ec..2f3f1d1 100644
--- a/service/model_resolve.go
+++ b/service/utils/model_resolve.go
@@ -1,4 +1,4 @@
-package service
+package utils
 
 import (
 	"context"
@@ -83,6 +83,7 @@ func substituteAPIPlaceholder(m *entity.ModelManage, key string) {
 // 配置字段取系统行；个人字段(apiKey/enabled/chatModel)取引用行；enabled 取 AND（系统停用=引用失效）。
 func mergeReferenceConfig(stub, sys *entity.ModelManage) *entity.ModelManage {
 	out := *sys
+	out.RefSystemModelId = stub.RefSystemModelId // 保留引用标记，调用侧据此判断引用行门禁（Id 已覆盖为系统模型 id）
 	out.ApiKey = stub.ApiKey
 	if stub.Enabled != nil {
 		out.Enabled = stub.Enabled
@@ -96,11 +97,11 @@ func mergeReferenceConfig(stub, sys *entity.ModelManage) *entity.ModelManage {
 	return &out
 }
 
-// mergeReferenceConfigForQuery 管理端 Get 查询展示用：以引用行为基底，把系统行的配置列合入，
+// MergeReferenceConfigForQuery 管理端 Get 查询展示用：以引用行为基底，把系统行的配置列合入，
 // 保留引用行自身 Id/RefSystemModelId/SystemModel/Creator/时间戳与个人字段(apiKey/enabled/chatModel)。
 // 与 mergeReferenceConfig 的区别：不替换 {apiKey}（Get 非引用行也不替换，展示模板），
 // enabled 不做 AND（展示引用行个人开关，调用时才按系统行生效状态门禁）。
-func mergeReferenceConfigForQuery(stub, sys *entity.ModelManage) *entity.ModelManage {
+func MergeReferenceConfigForQuery(stub, sys *entity.ModelManage) *entity.ModelManage {
 	out := *stub
 	out.BaseURL = sys.BaseURL
 	out.HttpMethod = sys.HttpMethod
@@ -124,10 +125,10 @@ func mergeReferenceConfigForQuery(stub, sys *entity.ModelManage) *entity.ModelMa
 	return &out
 }
 
-// resolveModelConfig 把请求命中的模型行解析为可执行配置：
+// ResolveModelConfig 把请求命中的模型行解析为可执行配置：
 // 引用行 → 系统行配置 + 引用行 apiKey（Id 覆盖为系统模型 id）；非引用行 → 原配置 + 自身 apiKey 替换占位。
 // 引用系统模型已删除 → 报错（调用方阻塞）。
-func resolveModelConfig(ctx context.Context, m *entity.ModelManage) (*entity.ModelManage, error) {
+func ResolveModelConfig(ctx context.Context, m *entity.ModelManage) (*entity.ModelManage, error) {
 	if m == nil {
 		return nil, nil
 	}