Merge branch 'release/custom-0.1.91' into release/custom-0.1.92

.github/workflows/backend-ci.yml

@@ -17,6 +17,7 @@ jobs:
           go-version-file: backend/go.mod
           check-latest: false
           cache: true
+          cache-dependency-path: backend/go.sum
       - name: Verify Go version
         run: |
           go version | grep -q 'go1.25.7'
@@ -36,6 +37,7 @@ jobs:
           go-version-file: backend/go.mod
           check-latest: false
           cache: true
+          cache-dependency-path: backend/go.sum
       - name: Verify Go version
         run: |
           go version | grep -q 'go1.25.7'

.gitignore

@@ -78,6 +78,7 @@ Desktop.ini
 # ===================
 tmp/
 temp/
+logs/
 *.tmp
 *.temp
 *.log
@@ -128,8 +129,15 @@ deploy/docker-compose.override.yml
 vite.config.js
 docs/*
 .serena/
+
+# ===================
+# 压测工具
+# ===================
+tools/loadtest/
+# Antigravity Manager
+Antigravity-Manager/
+antigravity_projectid_fix.patch
 .codex/
 frontend/coverage/
 aicodex
 output/
-

backend/cmd/server/VERSION

@@ -1 +1 @@
-0.1.88
+0.1.91.4

backend/ent/group.go

@@ -62,22 +62,24 @@ type Group struct {
 	SoraVideoPricePerRequestHd *float64 `json:"sora_video_price_per_request_hd,omitempty"`
 	// SoraStorageQuotaBytes holds the value of the "sora_storage_quota_bytes" field.
 	SoraStorageQuotaBytes int64 `json:"sora_storage_quota_bytes,omitempty"`
-	// 是否仅允许 Claude Code 客户端
+	// allow Claude Code client only
 	ClaudeCodeOnly bool `json:"claude_code_only,omitempty"`
-	// 非 Claude Code 请求降级使用的分组 ID
+	// fallback group for non-Claude-Code requests
 	FallbackGroupID *int64 `json:"fallback_group_id,omitempty"`
-	// 无效请求兜底使用的分组 ID
+	// fallback group for invalid request
 	FallbackGroupIDOnInvalidRequest *int64 `json:"fallback_group_id_on_invalid_request,omitempty"`
-	// 模型路由配置：模型模式 -> 优先账号ID列表
+	// model routing config: pattern -> account ids
 	ModelRouting map[string][]int64 `json:"model_routing,omitempty"`
-	// 是否启用模型路由配置
+	// whether model routing is enabled
 	ModelRoutingEnabled bool `json:"model_routing_enabled,omitempty"`
-	// 是否注入 MCP XML 调用协议提示词（仅 antigravity 平台）
+	// whether MCP XML prompt injection is enabled
 	McpXMLInject bool `json:"mcp_xml_inject,omitempty"`
-	// 支持的模型系列：claude, gemini_text, gemini_image
+	// supported model scopes: claude, gemini_text, gemini_image
 	SupportedModelScopes []string `json:"supported_model_scopes,omitempty"`
-	// 分组显示排序，数值越小越靠前
+	// group display order, lower comes first
 	SortOrder int `json:"sort_order,omitempty"`
+	// simulate claude usage as claude-max style (1h cache write)
+	SimulateClaudeMaxEnabled bool `json:"simulate_claude_max_enabled,omitempty"`
 	// Edges holds the relations/edges for other nodes in the graph.
 	// The values are being populated by the GroupQuery when eager-loading is set.
 	Edges        GroupEdges `json:"edges"`
@@ -186,7 +188,7 @@ func (*Group) scanValues(columns []string) ([]any, error) {
 		switch columns[i] {
 		case group.FieldModelRouting, group.FieldSupportedModelScopes:
 			values[i] = new([]byte)
-		case group.FieldIsExclusive, group.FieldClaudeCodeOnly, group.FieldModelRoutingEnabled, group.FieldMcpXMLInject:
+		case group.FieldIsExclusive, group.FieldClaudeCodeOnly, group.FieldModelRoutingEnabled, group.FieldMcpXMLInject, group.FieldSimulateClaudeMaxEnabled:
 			values[i] = new(sql.NullBool)
 		case group.FieldRateMultiplier, group.FieldDailyLimitUsd, group.FieldWeeklyLimitUsd, group.FieldMonthlyLimitUsd, group.FieldImagePrice1k, group.FieldImagePrice2k, group.FieldImagePrice4k, group.FieldSoraImagePrice360, group.FieldSoraImagePrice540, group.FieldSoraVideoPricePerRequest, group.FieldSoraVideoPricePerRequestHd:
 			values[i] = new(sql.NullFloat64)
@@ -415,6 +417,12 @@ func (_m *Group) assignValues(columns []string, values []any) error {
 			} else if value.Valid {
 				_m.SortOrder = int(value.Int64)
 			}
+		case group.FieldSimulateClaudeMaxEnabled:
+			if value, ok := values[i].(*sql.NullBool); !ok {
+				return fmt.Errorf("unexpected type %T for field simulate_claude_max_enabled", values[i])
+			} else if value.Valid {
+				_m.SimulateClaudeMaxEnabled = value.Bool
+			}
 		default:
 			_m.selectValues.Set(columns[i], values[i])
 		}
@@ -608,6 +616,9 @@ func (_m *Group) String() string {
 	builder.WriteString(", ")
 	builder.WriteString("sort_order=")
 	builder.WriteString(fmt.Sprintf("%v", _m.SortOrder))
+	builder.WriteString(", ")
+	builder.WriteString("simulate_claude_max_enabled=")
+	builder.WriteString(fmt.Sprintf("%v", _m.SimulateClaudeMaxEnabled))
 	builder.WriteByte(')')
 	return builder.String()
 }

backend/ent/group/group.go

@@ -75,6 +75,8 @@ const (
 	FieldSupportedModelScopes = "supported_model_scopes"
 	// FieldSortOrder holds the string denoting the sort_order field in the database.
 	FieldSortOrder = "sort_order"
+	// FieldSimulateClaudeMaxEnabled holds the string denoting the simulate_claude_max_enabled field in the database.
+	FieldSimulateClaudeMaxEnabled = "simulate_claude_max_enabled"
 	// EdgeAPIKeys holds the string denoting the api_keys edge name in mutations.
 	EdgeAPIKeys = "api_keys"
 	// EdgeRedeemCodes holds the string denoting the redeem_codes edge name in mutations.
@@ -180,6 +182,7 @@ var Columns = []string{
 	FieldMcpXMLInject,
 	FieldSupportedModelScopes,
 	FieldSortOrder,
+	FieldSimulateClaudeMaxEnabled,
 }
 
 var (
@@ -247,6 +250,8 @@ var (
 	DefaultSupportedModelScopes []string
 	// DefaultSortOrder holds the default value on creation for the "sort_order" field.
 	DefaultSortOrder int
+	// DefaultSimulateClaudeMaxEnabled holds the default value on creation for the "simulate_claude_max_enabled" field.
+	DefaultSimulateClaudeMaxEnabled bool
 )
 
 // OrderOption defines the ordering options for the Group queries.
@@ -397,6 +402,11 @@ func BySortOrder(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldSortOrder, opts...).ToFunc()
 }
 
+// BySimulateClaudeMaxEnabled orders the results by the simulate_claude_max_enabled field.
+func BySimulateClaudeMaxEnabled(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldSimulateClaudeMaxEnabled, opts...).ToFunc()
+}
+
 // ByAPIKeysCount orders the results by api_keys count.
 func ByAPIKeysCount(opts ...sql.OrderTermOption) OrderOption {
 	return func(s *sql.Selector) {

backend/ent/group/where.go

@@ -195,6 +195,11 @@ func SortOrder(v int) predicate.Group {
 	return predicate.Group(sql.FieldEQ(FieldSortOrder, v))
 }
 
+// SimulateClaudeMaxEnabled applies equality check predicate on the "simulate_claude_max_enabled" field. It's identical to SimulateClaudeMaxEnabledEQ.
+func SimulateClaudeMaxEnabled(v bool) predicate.Group {
+	return predicate.Group(sql.FieldEQ(FieldSimulateClaudeMaxEnabled, v))
+}
+
 // CreatedAtEQ applies the EQ predicate on the "created_at" field.
 func CreatedAtEQ(v time.Time) predicate.Group {
 	return predicate.Group(sql.FieldEQ(FieldCreatedAt, v))
@@ -1470,6 +1475,16 @@ func SortOrderLTE(v int) predicate.Group {
 	return predicate.Group(sql.FieldLTE(FieldSortOrder, v))
 }
 
+// SimulateClaudeMaxEnabledEQ applies the EQ predicate on the "simulate_claude_max_enabled" field.
+func SimulateClaudeMaxEnabledEQ(v bool) predicate.Group {
+	return predicate.Group(sql.FieldEQ(FieldSimulateClaudeMaxEnabled, v))
+}
+
+// SimulateClaudeMaxEnabledNEQ applies the NEQ predicate on the "simulate_claude_max_enabled" field.
+func SimulateClaudeMaxEnabledNEQ(v bool) predicate.Group {
+	return predicate.Group(sql.FieldNEQ(FieldSimulateClaudeMaxEnabled, v))
+}
+
 // HasAPIKeys applies the HasEdge predicate on the "api_keys" edge.
 func HasAPIKeys() predicate.Group {
 	return predicate.Group(func(s *sql.Selector) {

backend/ent/group_create.go

@@ -424,6 +424,20 @@ func (_c *GroupCreate) SetNillableSortOrder(v *int) *GroupCreate {
 	return _c
 }
 
+// SetSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field.
+func (_c *GroupCreate) SetSimulateClaudeMaxEnabled(v bool) *GroupCreate {
+	_c.mutation.SetSimulateClaudeMaxEnabled(v)
+	return _c
+}
+
+// SetNillableSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field if the given value is not nil.
+func (_c *GroupCreate) SetNillableSimulateClaudeMaxEnabled(v *bool) *GroupCreate {
+	if v != nil {
+		_c.SetSimulateClaudeMaxEnabled(*v)
+	}
+	return _c
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by IDs.
 func (_c *GroupCreate) AddAPIKeyIDs(ids ...int64) *GroupCreate {
 	_c.mutation.AddAPIKeyIDs(ids...)
@@ -613,6 +627,10 @@ func (_c *GroupCreate) defaults() error {
 		v := group.DefaultSortOrder
 		_c.mutation.SetSortOrder(v)
 	}
+	if _, ok := _c.mutation.SimulateClaudeMaxEnabled(); !ok {
+		v := group.DefaultSimulateClaudeMaxEnabled
+		_c.mutation.SetSimulateClaudeMaxEnabled(v)
+	}
 	return nil
 }
 
@@ -683,6 +701,9 @@ func (_c *GroupCreate) check() error {
 	if _, ok := _c.mutation.SortOrder(); !ok {
 		return &ValidationError{Name: "sort_order", err: errors.New(`ent: missing required field "Group.sort_order"`)}
 	}
+	if _, ok := _c.mutation.SimulateClaudeMaxEnabled(); !ok {
+		return &ValidationError{Name: "simulate_claude_max_enabled", err: errors.New(`ent: missing required field "Group.simulate_claude_max_enabled"`)}
+	}
 	return nil
 }
 
@@ -830,6 +851,10 @@ func (_c *GroupCreate) createSpec() (*Group, *sqlgraph.CreateSpec) {
 		_spec.SetField(group.FieldSortOrder, field.TypeInt, value)
 		_node.SortOrder = value
 	}
+	if value, ok := _c.mutation.SimulateClaudeMaxEnabled(); ok {
+		_spec.SetField(group.FieldSimulateClaudeMaxEnabled, field.TypeBool, value)
+		_node.SimulateClaudeMaxEnabled = value
+	}
 	if nodes := _c.mutation.APIKeysIDs(); len(nodes) > 0 {
 		edge := &sqlgraph.EdgeSpec{
 			Rel:     sqlgraph.O2M,
@@ -1520,6 +1545,18 @@ func (u *GroupUpsert) AddSortOrder(v int) *GroupUpsert {
 	return u
 }
 
+// SetSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field.
+func (u *GroupUpsert) SetSimulateClaudeMaxEnabled(v bool) *GroupUpsert {
+	u.Set(group.FieldSimulateClaudeMaxEnabled, v)
+	return u
+}
+
+// UpdateSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field to the value that was provided on create.
+func (u *GroupUpsert) UpdateSimulateClaudeMaxEnabled() *GroupUpsert {
+	u.SetExcluded(group.FieldSimulateClaudeMaxEnabled)
+	return u
+}
+
 // UpdateNewValues updates the mutable fields using the new values that were set on create.
 // Using this option is equivalent to using:
 //
@@ -2188,6 +2225,20 @@ func (u *GroupUpsertOne) UpdateSortOrder() *GroupUpsertOne {
 	})
 }
 
+// SetSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field.
+func (u *GroupUpsertOne) SetSimulateClaudeMaxEnabled(v bool) *GroupUpsertOne {
+	return u.Update(func(s *GroupUpsert) {
+		s.SetSimulateClaudeMaxEnabled(v)
+	})
+}
+
+// UpdateSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field to the value that was provided on create.
+func (u *GroupUpsertOne) UpdateSimulateClaudeMaxEnabled() *GroupUpsertOne {
+	return u.Update(func(s *GroupUpsert) {
+		s.UpdateSimulateClaudeMaxEnabled()
+	})
+}
+
 // Exec executes the query.
 func (u *GroupUpsertOne) Exec(ctx context.Context) error {
 	if len(u.create.conflict) == 0 {
@@ -3022,6 +3073,20 @@ func (u *GroupUpsertBulk) UpdateSortOrder() *GroupUpsertBulk {
 	})
 }
 
+// SetSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field.
+func (u *GroupUpsertBulk) SetSimulateClaudeMaxEnabled(v bool) *GroupUpsertBulk {
+	return u.Update(func(s *GroupUpsert) {
+		s.SetSimulateClaudeMaxEnabled(v)
+	})
+}
+
+// UpdateSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field to the value that was provided on create.
+func (u *GroupUpsertBulk) UpdateSimulateClaudeMaxEnabled() *GroupUpsertBulk {
+	return u.Update(func(s *GroupUpsert) {
+		s.UpdateSimulateClaudeMaxEnabled()
+	})
+}
+
 // Exec executes the query.
 func (u *GroupUpsertBulk) Exec(ctx context.Context) error {
 	if u.create.err != nil {

backend/ent/group_update.go

@@ -625,6 +625,20 @@ func (_u *GroupUpdate) AddSortOrder(v int) *GroupUpdate {
 	return _u
 }
 
+// SetSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field.
+func (_u *GroupUpdate) SetSimulateClaudeMaxEnabled(v bool) *GroupUpdate {
+	_u.mutation.SetSimulateClaudeMaxEnabled(v)
+	return _u
+}
+
+// SetNillableSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field if the given value is not nil.
+func (_u *GroupUpdate) SetNillableSimulateClaudeMaxEnabled(v *bool) *GroupUpdate {
+	if v != nil {
+		_u.SetSimulateClaudeMaxEnabled(*v)
+	}
+	return _u
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by IDs.
 func (_u *GroupUpdate) AddAPIKeyIDs(ids ...int64) *GroupUpdate {
 	_u.mutation.AddAPIKeyIDs(ids...)
@@ -1110,6 +1124,9 @@ func (_u *GroupUpdate) sqlSave(ctx context.Context) (_node int, err error) {
 	if value, ok := _u.mutation.AddedSortOrder(); ok {
 		_spec.AddField(group.FieldSortOrder, field.TypeInt, value)
 	}
+	if value, ok := _u.mutation.SimulateClaudeMaxEnabled(); ok {
+		_spec.SetField(group.FieldSimulateClaudeMaxEnabled, field.TypeBool, value)
+	}
 	if _u.mutation.APIKeysCleared() {
 		edge := &sqlgraph.EdgeSpec{
 			Rel:     sqlgraph.O2M,
@@ -2014,6 +2031,20 @@ func (_u *GroupUpdateOne) AddSortOrder(v int) *GroupUpdateOne {
 	return _u
 }
 
+// SetSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field.
+func (_u *GroupUpdateOne) SetSimulateClaudeMaxEnabled(v bool) *GroupUpdateOne {
+	_u.mutation.SetSimulateClaudeMaxEnabled(v)
+	return _u
+}
+
+// SetNillableSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field if the given value is not nil.
+func (_u *GroupUpdateOne) SetNillableSimulateClaudeMaxEnabled(v *bool) *GroupUpdateOne {
+	if v != nil {
+		_u.SetSimulateClaudeMaxEnabled(*v)
+	}
+	return _u
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by IDs.
 func (_u *GroupUpdateOne) AddAPIKeyIDs(ids ...int64) *GroupUpdateOne {
 	_u.mutation.AddAPIKeyIDs(ids...)
@@ -2529,6 +2560,9 @@ func (_u *GroupUpdateOne) sqlSave(ctx context.Context) (_node *Group, err error)
 	if value, ok := _u.mutation.AddedSortOrder(); ok {
 		_spec.AddField(group.FieldSortOrder, field.TypeInt, value)
 	}
+	if value, ok := _u.mutation.SimulateClaudeMaxEnabled(); ok {
+		_spec.SetField(group.FieldSimulateClaudeMaxEnabled, field.TypeBool, value)
+	}
 	if _u.mutation.APIKeysCleared() {
 		edge := &sqlgraph.EdgeSpec{
 			Rel:     sqlgraph.O2M,

backend/ent/migrate/schema.go

@@ -407,6 +407,7 @@ var (
 		{Name: "mcp_xml_inject", Type: field.TypeBool, Default: true},
 		{Name: "supported_model_scopes", Type: field.TypeJSON, SchemaType: map[string]string{"postgres": "jsonb"}},
 		{Name: "sort_order", Type: field.TypeInt, Default: 0},
+		{Name: "simulate_claude_max_enabled", Type: field.TypeBool, Default: false},
 	}
 	// GroupsTable holds the schema information for the "groups" table.
 	GroupsTable = &schema.Table{

backend/ent/mutation.go

@@ -8196,6 +8196,7 @@ type GroupMutation struct {
 	appendsupported_model_scopes            []string
 	sort_order                              *int
 	addsort_order                           *int
+	simulate_claude_max_enabled             *bool
 	clearedFields                           map[string]struct{}
 	api_keys                                map[int64]struct{}
 	removedapi_keys                         map[int64]struct{}
@@ -9940,6 +9941,42 @@ func (m *GroupMutation) ResetSortOrder() {
 	m.addsort_order = nil
 }
 
+// SetSimulateClaudeMaxEnabled sets the "simulate_claude_max_enabled" field.
+func (m *GroupMutation) SetSimulateClaudeMaxEnabled(b bool) {
+	m.simulate_claude_max_enabled = &b
+}
+
+// SimulateClaudeMaxEnabled returns the value of the "simulate_claude_max_enabled" field in the mutation.
+func (m *GroupMutation) SimulateClaudeMaxEnabled() (r bool, exists bool) {
+	v := m.simulate_claude_max_enabled
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldSimulateClaudeMaxEnabled returns the old "simulate_claude_max_enabled" field's value of the Group entity.
+// If the Group object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *GroupMutation) OldSimulateClaudeMaxEnabled(ctx context.Context) (v bool, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldSimulateClaudeMaxEnabled is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldSimulateClaudeMaxEnabled requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldSimulateClaudeMaxEnabled: %w", err)
+	}
+	return oldValue.SimulateClaudeMaxEnabled, nil
+}
+
+// ResetSimulateClaudeMaxEnabled resets all changes to the "simulate_claude_max_enabled" field.
+func (m *GroupMutation) ResetSimulateClaudeMaxEnabled() {
+	m.simulate_claude_max_enabled = nil
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by ids.
 func (m *GroupMutation) AddAPIKeyIDs(ids ...int64) {
 	if m.api_keys == nil {
@@ -10389,6 +10426,9 @@ func (m *GroupMutation) Fields() []string {
 	if m.sort_order != nil {
 		fields = append(fields, group.FieldSortOrder)
 	}
+	if m.simulate_claude_max_enabled != nil {
+		fields = append(fields, group.FieldSimulateClaudeMaxEnabled)
+	}
 	return fields
 }
 
@@ -10457,6 +10497,8 @@ func (m *GroupMutation) Field(name string) (ent.Value, bool) {
 		return m.SupportedModelScopes()
 	case group.FieldSortOrder:
 		return m.SortOrder()
+	case group.FieldSimulateClaudeMaxEnabled:
+		return m.SimulateClaudeMaxEnabled()
 	}
 	return nil, false
 }
@@ -10526,6 +10568,8 @@ func (m *GroupMutation) OldField(ctx context.Context, name string) (ent.Value, e
 		return m.OldSupportedModelScopes(ctx)
 	case group.FieldSortOrder:
 		return m.OldSortOrder(ctx)
+	case group.FieldSimulateClaudeMaxEnabled:
+		return m.OldSimulateClaudeMaxEnabled(ctx)
 	}
 	return nil, fmt.Errorf("unknown Group field %s", name)
 }
@@ -10745,6 +10789,13 @@ func (m *GroupMutation) SetField(name string, value ent.Value) error {
 		}
 		m.SetSortOrder(v)
 		return nil
+	case group.FieldSimulateClaudeMaxEnabled:
+		v, ok := value.(bool)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetSimulateClaudeMaxEnabled(v)
+		return nil
 	}
 	return fmt.Errorf("unknown Group field %s", name)
 }
@@ -11172,6 +11223,9 @@ func (m *GroupMutation) ResetField(name string) error {
 	case group.FieldSortOrder:
 		m.ResetSortOrder()
 		return nil
+	case group.FieldSimulateClaudeMaxEnabled:
+		m.ResetSimulateClaudeMaxEnabled()
+		return nil
 	}
 	return fmt.Errorf("unknown Group field %s", name)
 }

backend/ent/runtime/runtime.go

@@ -447,6 +447,10 @@ func init() {
 	groupDescSortOrder := groupFields[26].Descriptor()
 	// group.DefaultSortOrder holds the default value on creation for the sort_order field.
 	group.DefaultSortOrder = groupDescSortOrder.Default.(int)
+	// groupDescSimulateClaudeMaxEnabled is the schema descriptor for simulate_claude_max_enabled field.
+	groupDescSimulateClaudeMaxEnabled := groupFields[27].Descriptor()
+	// group.DefaultSimulateClaudeMaxEnabled holds the default value on creation for the simulate_claude_max_enabled field.
+	group.DefaultSimulateClaudeMaxEnabled = groupDescSimulateClaudeMaxEnabled.Default.(bool)
 	idempotencyrecordMixin := schema.IdempotencyRecord{}.Mixin()
 	idempotencyrecordMixinFields0 := idempotencyrecordMixin[0].Fields()
 	_ = idempotencyrecordMixinFields0

backend/ent/schema/group.go

@@ -33,8 +33,6 @@ func (Group) Mixin() []ent.Mixin {
 
 func (Group) Fields() []ent.Field {
 	return []ent.Field{
-		// 唯一约束通过部分索引实现（WHERE deleted_at IS NULL），支持软删除后重用
-		// 见迁移文件 016_soft_delete_partial_unique_indexes.sql
 		field.String("name").
 			MaxLen(100).
 			NotEmpty(),
@@ -51,7 +49,6 @@ func (Group) Fields() []ent.Field {
 			MaxLen(20).
 			Default(domain.StatusActive),
 
-		// Subscription-related fields (added by migration 003)
 		field.String("platform").
 			MaxLen(50).
 			Default(domain.PlatformAnthropic),
@@ -73,7 +70,6 @@ func (Group) Fields() []ent.Field {
 		field.Int("default_validity_days").
 			Default(30),
 
-		// 图片生成计费配置（antigravity 和 gemini 平台使用）
 		field.Float("image_price_1k").
 			Optional().
 			Nillable().
@@ -87,7 +83,6 @@ func (Group) Fields() []ent.Field {
 			Nillable().
 			SchemaType(map[string]string{dialect.Postgres: "decimal(20,8)"}),
 
-		// Sora 按次计费配置（阶段 1）
 		field.Float("sora_image_price_360").
 			Optional().
 			Nillable().
@@ -109,45 +104,41 @@ func (Group) Fields() []ent.Field {
 		field.Int64("sora_storage_quota_bytes").
 			Default(0),
 
-		// Claude Code 客户端限制 (added by migration 029)
 		field.Bool("claude_code_only").
 			Default(false).
-			Comment("是否仅允许 Claude Code 客户端"),
+			Comment("allow Claude Code client only"),
 		field.Int64("fallback_group_id").
 			Optional().
 			Nillable().
-			Comment("非 Claude Code 请求降级使用的分组 ID"),
+			Comment("fallback group for non-Claude-Code requests"),
 		field.Int64("fallback_group_id_on_invalid_request").
 			Optional().
 			Nillable().
-			Comment("无效请求兜底使用的分组 ID"),
+			Comment("fallback group for invalid request"),
 
-		// 模型路由配置 (added by migration 040)
 		field.JSON("model_routing", map[string][]int64{}).
 			Optional().
 			SchemaType(map[string]string{dialect.Postgres: "jsonb"}).
-			Comment("模型路由配置：模型模式 -> 优先账号ID列表"),
-
-		// 模型路由开关 (added by migration 041)
+			Comment("model routing config: pattern -> account ids"),
 		field.Bool("model_routing_enabled").
 			Default(false).
-			Comment("是否启用模型路由配置"),
+			Comment("whether model routing is enabled"),
 
-		// MCP XML 协议注入开关 (added by migration 042)
 		field.Bool("mcp_xml_inject").
 			Default(true).
-			Comment("是否注入 MCP XML 调用协议提示词（仅 antigravity 平台）"),
+			Comment("whether MCP XML prompt injection is enabled"),
 
-		// 支持的模型系列 (added by migration 046)
 		field.JSON("supported_model_scopes", []string{}).
 			Default([]string{"claude", "gemini_text", "gemini_image"}).
 			SchemaType(map[string]string{dialect.Postgres: "jsonb"}).
-			Comment("支持的模型系列：claude, gemini_text, gemini_image"),
+			Comment("supported model scopes: claude, gemini_text, gemini_image"),
 
-		// 分组排序 (added by migration 052)
 		field.Int("sort_order").
 			Default(0).
-			Comment("分组显示排序，数值越小越靠前"),
+			Comment("group display order, lower comes first"),
+		field.Bool("simulate_claude_max_enabled").
+			Default(false).
+			Comment("simulate claude usage as claude-max style (1h cache write)"),
 	}
 }
 
@@ -163,14 +154,11 @@ func (Group) Edges() []ent.Edge {
 		edge.From("allowed_users", User.Type).
 			Ref("allowed_groups").
 			Through("user_allowed_groups", UserAllowedGroup.Type),
-		// 注意：fallback_group_id 直接作为字段使用，不定义 edge
-		// 这样允许多个分组指向同一个降级分组（M2O 关系）
 	}
 }
 
 func (Group) Indexes() []ent.Index {
 	return []ent.Index{
-		// name 字段已在 Fields() 中声明 Unique()，无需重复索引
 		index.Fields("status"),
 		index.Fields("platform"),
 		index.Fields("subscription_type"),

backend/go.mod

@@ -89,6 +89,7 @@ require (
 	github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc // indirect
 	github.com/dgryski/go-rendezvous v0.0.0-20200823014737-9f7001d12a5f // indirect
 	github.com/distribution/reference v0.6.0 // indirect
+	github.com/dlclark/regexp2 v1.10.0 // indirect
 	github.com/docker/docker v28.5.1+incompatible // indirect
 	github.com/docker/go-connections v0.6.0 // indirect
 	github.com/docker/go-units v0.5.0 // indirect
@@ -140,6 +141,8 @@ require (
 	github.com/opencontainers/image-spec v1.1.1 // indirect
 	github.com/pelletier/go-toml/v2 v2.2.2 // indirect
 	github.com/pkg/errors v0.9.1 // indirect
+	github.com/pkoukk/tiktoken-go v0.1.8 // indirect
+	github.com/pkoukk/tiktoken-go-loader v0.0.2 // indirect
 	github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2 // indirect
 	github.com/power-devops/perfstat v0.0.0-20210106213030-5aafc221ea8c // indirect
 	github.com/quic-go/qpack v0.6.0 // indirect

backend/internal/handler/admin/account_handler.go

@@ -1155,6 +1155,12 @@ func (h *AccountHandler) BulkUpdate(c *gin.Context) {
 			c.JSON(409, gin.H{
 				"error":   "mixed_channel_warning",
 				"message": mixedErr.Error(),
+				"details": gin.H{
+					"group_id":         mixedErr.GroupID,
+					"group_name":       mixedErr.GroupName,
+					"current_platform": mixedErr.CurrentPlatform,
+					"other_platform":   mixedErr.OtherPlatform,
+				},
 			})
 			return
 		}

backend/internal/handler/admin/account_handler_mixed_channel_test.go

@@ -111,7 +111,7 @@ func TestAccountHandlerCreateMixedChannelConflictSimplifiedResponse(t *testing.T
 	var resp map[string]any
 	require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &resp))
 	require.Equal(t, "mixed_channel_warning", resp["error"])
-	require.Contains(t, resp["message"], "mixed_channel_warning")
+	require.Contains(t, resp["message"], "claude-max")
 	_, hasDetails := resp["details"]
 	_, hasRequireConfirmation := resp["require_confirmation"]
 	require.False(t, hasDetails)
@@ -140,7 +140,7 @@ func TestAccountHandlerUpdateMixedChannelConflictSimplifiedResponse(t *testing.T
 	var resp map[string]any
 	require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &resp))
 	require.Equal(t, "mixed_channel_warning", resp["error"])
-	require.Contains(t, resp["message"], "mixed_channel_warning")
+	require.Contains(t, resp["message"], "claude-max")
 	_, hasDetails := resp["details"]
 	_, hasRequireConfirmation := resp["require_confirmation"]
 	require.False(t, hasDetails)

backend/internal/handler/admin/group_handler.go

@@ -46,9 +46,10 @@ type CreateGroupRequest struct {
 	FallbackGroupID                 *int64   `json:"fallback_group_id"`
 	FallbackGroupIDOnInvalidRequest *int64   `json:"fallback_group_id_on_invalid_request"`
 	// 模型路由配置（仅 anthropic 平台使用）
-	ModelRouting        map[string][]int64 `json:"model_routing"`
-	ModelRoutingEnabled bool               `json:"model_routing_enabled"`
-	MCPXMLInject        *bool              `json:"mcp_xml_inject"`
+	ModelRouting             map[string][]int64 `json:"model_routing"`
+	ModelRoutingEnabled      bool               `json:"model_routing_enabled"`
+	MCPXMLInject             *bool              `json:"mcp_xml_inject"`
+	SimulateClaudeMaxEnabled *bool              `json:"simulate_claude_max_enabled"`
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string `json:"supported_model_scopes"`
 	// Sora 存储配额
@@ -81,9 +82,10 @@ type UpdateGroupRequest struct {
 	FallbackGroupID                 *int64   `json:"fallback_group_id"`
 	FallbackGroupIDOnInvalidRequest *int64   `json:"fallback_group_id_on_invalid_request"`
 	// 模型路由配置（仅 anthropic 平台使用）
-	ModelRouting        map[string][]int64 `json:"model_routing"`
-	ModelRoutingEnabled *bool              `json:"model_routing_enabled"`
-	MCPXMLInject        *bool              `json:"mcp_xml_inject"`
+	ModelRouting             map[string][]int64 `json:"model_routing"`
+	ModelRoutingEnabled      *bool              `json:"model_routing_enabled"`
+	MCPXMLInject             *bool              `json:"mcp_xml_inject"`
+	SimulateClaudeMaxEnabled *bool              `json:"simulate_claude_max_enabled"`
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes *[]string `json:"supported_model_scopes"`
 	// Sora 存储配额
@@ -201,6 +203,7 @@ func (h *GroupHandler) Create(c *gin.Context) {
 		ModelRouting:                    req.ModelRouting,
 		ModelRoutingEnabled:             req.ModelRoutingEnabled,
 		MCPXMLInject:                    req.MCPXMLInject,
+		SimulateClaudeMaxEnabled:        req.SimulateClaudeMaxEnabled,
 		SupportedModelScopes:            req.SupportedModelScopes,
 		SoraStorageQuotaBytes:           req.SoraStorageQuotaBytes,
 		CopyAccountsFromGroupIDs:        req.CopyAccountsFromGroupIDs,
@@ -252,6 +255,7 @@ func (h *GroupHandler) Update(c *gin.Context) {
 		ModelRouting:                    req.ModelRouting,
 		ModelRoutingEnabled:             req.ModelRoutingEnabled,
 		MCPXMLInject:                    req.MCPXMLInject,
+		SimulateClaudeMaxEnabled:        req.SimulateClaudeMaxEnabled,
 		SupportedModelScopes:            req.SupportedModelScopes,
 		SoraStorageQuotaBytes:           req.SoraStorageQuotaBytes,
 		CopyAccountsFromGroupIDs:        req.CopyAccountsFromGroupIDs,

backend/internal/handler/dto/mappers.go

@@ -122,13 +122,14 @@ func GroupFromServiceAdmin(g *service.Group) *AdminGroup {
 		return nil
 	}
 	out := &AdminGroup{
-		Group:                groupFromServiceBase(g),
-		ModelRouting:         g.ModelRouting,
-		ModelRoutingEnabled:  g.ModelRoutingEnabled,
-		MCPXMLInject:         g.MCPXMLInject,
-		SupportedModelScopes: g.SupportedModelScopes,
-		AccountCount:         g.AccountCount,
-		SortOrder:            g.SortOrder,
+		Group:                    groupFromServiceBase(g),
+		ModelRouting:             g.ModelRouting,
+		ModelRoutingEnabled:      g.ModelRoutingEnabled,
+		MCPXMLInject:             g.MCPXMLInject,
+		SimulateClaudeMaxEnabled: g.SimulateClaudeMaxEnabled,
+		SupportedModelScopes:     g.SupportedModelScopes,
+		AccountCount:             g.AccountCount,
+		SortOrder:                g.SortOrder,
 	}
 	if len(g.AccountGroups) > 0 {
 		out.AccountGroups = make([]AccountGroup, 0, len(g.AccountGroups))
@@ -249,6 +250,12 @@ func AccountFromServiceShallow(a *service.Account) *Account {
 		}
 	}
 
+	// 客户端亲和调度（适用于 Anthropic/Antigravity 平台）
+	if a.IsClientAffinityEnabled() {
+		enabled := true
+		out.ClientAffinityEnabled = &enabled
+	}
+
 	// 提取 API Key 账号配额限制（仅 apikey 类型有效）
 	if a.Type == service.AccountTypeAPIKey {
 		if limit := a.GetQuotaLimit(); limit > 0 {

backend/internal/handler/dto/types.go

@@ -111,6 +111,8 @@ type AdminGroup struct {
 
 	// MCP XML 协议注入（仅 antigravity 平台使用）
 	MCPXMLInject bool `json:"mcp_xml_inject"`
+	// Claude usage 模拟开关（仅管理员可见）
+	SimulateClaudeMaxEnabled bool `json:"simulate_claude_max_enabled"`
 
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string       `json:"supported_model_scopes"`
@@ -186,6 +188,10 @@ type Account struct {
 	CacheTTLOverrideEnabled *bool   `json:"cache_ttl_override_enabled,omitempty"`
 	CacheTTLOverrideTarget  *string `json:"cache_ttl_override_target,omitempty"`
 
+	// 客户端亲和调度（仅 Anthropic/Antigravity 平台有效）
+	// 启用后新会话会优先调度到客户端之前使用过的账号
+	ClientAffinityEnabled *bool `json:"client_affinity_enabled,omitempty"`
+
 	// API Key 账号配额限制
 	QuotaLimit *float64 `json:"quota_limit,omitempty"`
 	QuotaUsed  *float64 `json:"quota_used,omitempty"`

backend/internal/handler/gateway_handler.go

@@ -439,6 +439,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 			h.submitUsageRecordTask(func(ctx context.Context) {
 				if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{
 					Result:            result,
+					ParsedRequest:     parsedReq,
 					APIKey:            apiKey,
 					User:              apiKey.User,
 					Account:           account,
@@ -630,6 +631,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 			// ===== 用户消息串行队列 END =====
 
 			// 转发请求 - 根据账号平台分流
+			c.Set("parsed_request", parsedReq)
 			var result *service.ForwardResult
 			requestCtx := c.Request.Context()
 			if fs.SwitchCount > 0 {
@@ -734,6 +736,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 			h.submitUsageRecordTask(func(ctx context.Context) {
 				if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{
 					Result:            result,
+					ParsedRequest:     parsedReq,
 					APIKey:            currentAPIKey,
 					User:              currentAPIKey.User,
 					Account:           account,

backend/internal/pkg/antigravity/stream_transformer.go

@@ -18,6 +18,9 @@ const (
 	BlockTypeFunction
 )
 
+// UsageMapHook is a callback that can modify usage data before it's emitted in SSE events.
+type UsageMapHook func(usageMap map[string]any)
+
 // StreamingProcessor 流式响应处理器
 type StreamingProcessor struct {
 	blockType         BlockType
@@ -30,6 +33,7 @@ type StreamingProcessor struct {
 	originalModel     string
 	webSearchQueries  []string
 	groundingChunks   []GeminiGroundingChunk
+	usageMapHook      UsageMapHook
 
 	// 累计 usage
 	inputTokens     int
@@ -45,6 +49,25 @@ func NewStreamingProcessor(originalModel string) *StreamingProcessor {
 	}
 }
 
+// SetUsageMapHook sets an optional hook that modifies usage maps before they are emitted.
+func (p *StreamingProcessor) SetUsageMapHook(fn UsageMapHook) {
+	p.usageMapHook = fn
+}
+
+func usageToMap(u ClaudeUsage) map[string]any {
+	m := map[string]any{
+		"input_tokens":  u.InputTokens,
+		"output_tokens": u.OutputTokens,
+	}
+	if u.CacheCreationInputTokens > 0 {
+		m["cache_creation_input_tokens"] = u.CacheCreationInputTokens
+	}
+	if u.CacheReadInputTokens > 0 {
+		m["cache_read_input_tokens"] = u.CacheReadInputTokens
+	}
+	return m
+}
+
 // ProcessLine 处理 SSE 行，返回 Claude SSE 事件
 func (p *StreamingProcessor) ProcessLine(line string) []byte {
 	line = strings.TrimSpace(line)
@@ -158,6 +181,13 @@ func (p *StreamingProcessor) emitMessageStart(v1Resp *V1InternalResponse) []byte
 		responseID = "msg_" + generateRandomID()
 	}
 
+	var usageValue any = usage
+	if p.usageMapHook != nil {
+		usageMap := usageToMap(usage)
+		p.usageMapHook(usageMap)
+		usageValue = usageMap
+	}
+
 	message := map[string]any{
 		"id":            responseID,
 		"type":          "message",
@@ -166,7 +196,7 @@ func (p *StreamingProcessor) emitMessageStart(v1Resp *V1InternalResponse) []byte
 		"model":         p.originalModel,
 		"stop_reason":   nil,
 		"stop_sequence": nil,
-		"usage":         usage,
+		"usage":         usageValue,
 	}
 
 	event := map[string]any{
@@ -477,13 +507,20 @@ func (p *StreamingProcessor) emitFinish(finishReason string) []byte {
 		CacheReadInputTokens: p.cacheReadTokens,
 	}
 
+	var usageValue any = usage
+	if p.usageMapHook != nil {
+		usageMap := usageToMap(usage)
+		p.usageMapHook(usageMap)
+		usageValue = usageMap
+	}
+
 	deltaEvent := map[string]any{
 		"type": "message_delta",
 		"delta": map[string]any{
 			"stop_reason":   stopReason,
 			"stop_sequence": nil,
 		},
-		"usage": usage,
+		"usage": usageValue,
 	}
 
 	_, _ = result.Write(p.formatSSE("message_delta", deltaEvent))

backend/internal/repository/api_key_repo.go

@@ -164,6 +164,7 @@ func (r *apiKeyRepository) GetByKeyForAuth(ctx context.Context, key string) (*se
 				group.FieldModelRoutingEnabled,
 				group.FieldModelRouting,
 				group.FieldMcpXMLInject,
+				group.FieldSimulateClaudeMaxEnabled,
 				group.FieldSupportedModelScopes,
 			)
 		}).
@@ -617,6 +618,7 @@ func groupEntityToService(g *dbent.Group) *service.Group {
 		ModelRouting:                    g.ModelRouting,
 		ModelRoutingEnabled:             g.ModelRoutingEnabled,
 		MCPXMLInject:                    g.McpXMLInject,
+		SimulateClaudeMaxEnabled:        g.SimulateClaudeMaxEnabled,
 		SupportedModelScopes:            g.SupportedModelScopes,
 		SortOrder:                       g.SortOrder,
 		CreatedAt:                       g.CreatedAt,

backend/internal/repository/gateway_cache.go

@@ -3,13 +3,17 @@ package repository
 import (
 	"context"
 	"fmt"
+	"strconv"
 	"time"
 
 	"github.com/Wei-Shaw/sub2api/internal/service"
 	"github.com/redis/go-redis/v9"
 )
 
-const stickySessionPrefix = "sticky_session:"
+const (
+	stickySessionPrefix  = "sticky_session:"
+	clientAffinityPrefix = "client_affinity:"
+)
 
 type gatewayCache struct {
 	rdb *redis.Client
@@ -51,3 +55,70 @@ func (c *gatewayCache) DeleteSessionAccountID(ctx context.Context, groupID int64
 	key := buildSessionKey(groupID, sessionHash)
 	return c.rdb.Del(ctx, key).Err()
 }
+
+// buildAffinityKey 构建客户端亲和 key
+// 格式: client_affinity:{groupID}:{clientID}
+func buildAffinityKey(groupID int64, clientID string) string {
+	return fmt.Sprintf("%s%d:%s", clientAffinityPrefix, groupID, clientID)
+}
+
+// getAffinityScript: 清理过期成员后返回亲和账号列表（按最近使用降序）
+// KEYS[1] = client_affinity:{groupID}:{clientID}
+// ARGV[1] = 过期阈值时间戳 (now - ttl)
+var getAffinityScript = redis.NewScript(`
+redis.call('ZREMRANGEBYSCORE', KEYS[1], '-inf', ARGV[1])
+return redis.call('ZREVRANGE', KEYS[1], 0, -1)
+`)
+
+// updateAffinityScript: 清理过期成员、添加/更新亲和关系、刷新 TTL
+// KEYS[1] = client_affinity:{groupID}:{clientID}
+// ARGV[1] = 当前时间戳
+// ARGV[2] = TTL 秒数
+// ARGV[3] = accountID
+// ARGV[4] = 过期阈值时间戳 (now - ttl)
+var updateAffinityScript = redis.NewScript(`
+redis.call('ZREMRANGEBYSCORE', KEYS[1], '-inf', ARGV[4])
+redis.call('ZADD', KEYS[1], ARGV[1], ARGV[3])
+redis.call('EXPIRE', KEYS[1], ARGV[2])
+return 1
+`)
+
+func (c *gatewayCache) GetClientAffinityAccounts(ctx context.Context, groupID int64, clientID string, ttl time.Duration) ([]int64, error) {
+	key := buildAffinityKey(groupID, clientID)
+	now := time.Now().Unix()
+	expireThreshold := now - int64(ttl.Seconds())
+
+	result, err := getAffinityScript.Run(ctx, c.rdb, []string{key}, expireThreshold).StringSlice()
+	if err != nil {
+		if err == redis.Nil {
+			return nil, nil
+		}
+		return nil, err
+	}
+
+	accountIDs := make([]int64, 0, len(result))
+	for _, s := range result {
+		id, err := strconv.ParseInt(s, 10, 64)
+		if err != nil {
+			continue
+		}
+		accountIDs = append(accountIDs, id)
+	}
+	return accountIDs, nil
+}
+
+func (c *gatewayCache) UpdateClientAffinity(ctx context.Context, groupID int64, clientID string, accountID int64, ttl time.Duration) error {
+	key := buildAffinityKey(groupID, clientID)
+	now := time.Now().Unix()
+	ttlSeconds := int64(ttl.Seconds())
+	expireThreshold := now - ttlSeconds
+
+	return updateAffinityScript.Run(ctx, c.rdb, []string{key},
+		now, ttlSeconds, accountID, expireThreshold,
+	).Err()
+}
+
+func (c *gatewayCache) RemoveClientAffinity(ctx context.Context, groupID int64, clientID string, accountID int64) error {
+	key := buildAffinityKey(groupID, clientID)
+	return c.rdb.ZRem(ctx, key, accountID).Err()
+}

backend/internal/repository/group_repo.go

@@ -59,7 +59,8 @@ func (r *groupRepository) Create(ctx context.Context, groupIn *service.Group) er
 		SetNillableFallbackGroupIDOnInvalidRequest(groupIn.FallbackGroupIDOnInvalidRequest).
 		SetModelRoutingEnabled(groupIn.ModelRoutingEnabled).
 		SetMcpXMLInject(groupIn.MCPXMLInject).
-		SetSoraStorageQuotaBytes(groupIn.SoraStorageQuotaBytes)
+		SetSoraStorageQuotaBytes(groupIn.SoraStorageQuotaBytes).
+		SetSimulateClaudeMaxEnabled(groupIn.SimulateClaudeMaxEnabled)
 
 	// 设置模型路由配置
 	if groupIn.ModelRouting != nil {
@@ -125,7 +126,8 @@ func (r *groupRepository) Update(ctx context.Context, groupIn *service.Group) er
 		SetClaudeCodeOnly(groupIn.ClaudeCodeOnly).
 		SetModelRoutingEnabled(groupIn.ModelRoutingEnabled).
 		SetMcpXMLInject(groupIn.MCPXMLInject).
-		SetSoraStorageQuotaBytes(groupIn.SoraStorageQuotaBytes)
+		SetSoraStorageQuotaBytes(groupIn.SoraStorageQuotaBytes).
+		SetSimulateClaudeMaxEnabled(groupIn.SimulateClaudeMaxEnabled)
 
 	// 显式处理可空字段：nil 需要 clear，非 nil 需要 set。
 	if groupIn.DailyLimitUSD != nil {

backend/internal/repository/usage_log_repo.go

@@ -1870,7 +1870,7 @@ func (r *usageLogRepository) GetGroupStatsWithFilters(ctx context.Context, start
 	query := `
 		SELECT
 			COALESCE(ul.group_id, 0) as group_id,
-			COALESCE(g.name, '') as group_name,
+			COALESCE(g.name, '(无分组)') as group_name,
 			COUNT(*) as requests,
 			COALESCE(SUM(ul.input_tokens + ul.output_tokens + ul.cache_creation_tokens + ul.cache_read_tokens), 0) as total_tokens,
 			COALESCE(SUM(ul.total_cost), 0) as cost,

backend/internal/service/account.go

@@ -1099,6 +1099,24 @@ func (a *Account) IsSessionIDMaskingEnabled() bool {
 	return false
 }
 
+// IsClientAffinityEnabled 检查是否启用客户端亲和调度
+// 仅适用于 Anthropic 平台账号（包括 anthropic 和 antigravity 平台）
+// 启用后，新会话会优先调度到之前使用过的账号
+func (a *Account) IsClientAffinityEnabled() bool {
+	if a.Platform != PlatformAnthropic && a.Platform != PlatformAntigravity {
+		return false
+	}
+	if a.Extra == nil {
+		return false
+	}
+	if v, ok := a.Extra["client_affinity_enabled"]; ok {
+		if enabled, ok := v.(bool); ok {
+			return enabled
+		}
+	}
+	return false
+}
+
 // IsCacheTTLOverrideEnabled 检查是否启用缓存 TTL 强制替换
 // 仅适用于 Anthropic OAuth/SetupToken 类型账号
 // 启用后将所有 cache creation tokens 归入指定的 TTL 类型（5m 或 1h）

backend/internal/service/admin_service.go

@@ -138,9 +138,10 @@ type CreateGroupInput struct {
 	// 无效请求兜底分组 ID（仅 anthropic 平台使用）
 	FallbackGroupIDOnInvalidRequest *int64
 	// 模型路由配置（仅 anthropic 平台使用）
-	ModelRouting        map[string][]int64
-	ModelRoutingEnabled bool // 是否启用模型路由
-	MCPXMLInject        *bool
+	ModelRouting             map[string][]int64
+	ModelRoutingEnabled      bool // 是否启用模型路由
+	MCPXMLInject             *bool
+	SimulateClaudeMaxEnabled *bool
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string
 	// Sora 存储配额
@@ -174,9 +175,10 @@ type UpdateGroupInput struct {
 	// 无效请求兜底分组 ID（仅 anthropic 平台使用）
 	FallbackGroupIDOnInvalidRequest *int64
 	// 模型路由配置（仅 anthropic 平台使用）
-	ModelRouting        map[string][]int64
-	ModelRoutingEnabled *bool // 是否启用模型路由
-	MCPXMLInject        *bool
+	ModelRouting             map[string][]int64
+	ModelRoutingEnabled      *bool // 是否启用模型路由
+	MCPXMLInject             *bool
+	SimulateClaudeMaxEnabled *bool
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes *[]string
 	// Sora 存储配额
@@ -357,6 +359,10 @@ type ProxyExitInfoProber interface {
 	ProbeProxy(ctx context.Context, proxyURL string) (*ProxyExitInfo, int64, error)
 }
 
+type groupExistenceBatchReader interface {
+	ExistsByIDs(ctx context.Context, ids []int64) (map[int64]bool, error)
+}
+
 type proxyQualityTarget struct {
 	Target          string
 	URL             string
@@ -432,10 +438,6 @@ type userGroupRateBatchReader interface {
 	GetByUserIDs(ctx context.Context, userIDs []int64) (map[int64]map[int64]float64, error)
 }
 
-type groupExistenceBatchReader interface {
-	ExistsByIDs(ctx context.Context, ids []int64) (map[int64]bool, error)
-}
-
 // NewAdminService creates a new AdminService
 func NewAdminService(
 	userRepo UserRepository,
@@ -851,6 +853,13 @@ func (s *adminServiceImpl) CreateGroup(ctx context.Context, input *CreateGroupIn
 	if input.MCPXMLInject != nil {
 		mcpXMLInject = *input.MCPXMLInject
 	}
+	simulateClaudeMaxEnabled := false
+	if input.SimulateClaudeMaxEnabled != nil {
+		if platform != PlatformAnthropic && *input.SimulateClaudeMaxEnabled {
+			return nil, fmt.Errorf("simulate_claude_max_enabled only supported for anthropic groups")
+		}
+		simulateClaudeMaxEnabled = *input.SimulateClaudeMaxEnabled
+	}
 
 	// 如果指定了复制账号的源分组，先获取账号 ID 列表
 	var accountIDsToCopy []int64
@@ -907,6 +916,7 @@ func (s *adminServiceImpl) CreateGroup(ctx context.Context, input *CreateGroupIn
 		FallbackGroupIDOnInvalidRequest: fallbackOnInvalidRequest,
 		ModelRouting:                    input.ModelRouting,
 		MCPXMLInject:                    mcpXMLInject,
+		SimulateClaudeMaxEnabled:        simulateClaudeMaxEnabled,
 		SupportedModelScopes:            input.SupportedModelScopes,
 		SoraStorageQuotaBytes:           input.SoraStorageQuotaBytes,
 	}
@@ -1116,6 +1126,15 @@ func (s *adminServiceImpl) UpdateGroup(ctx context.Context, id int64, input *Upd
 	if input.MCPXMLInject != nil {
 		group.MCPXMLInject = *input.MCPXMLInject
 	}
+	if input.SimulateClaudeMaxEnabled != nil {
+		if group.Platform != PlatformAnthropic && *input.SimulateClaudeMaxEnabled {
+			return nil, fmt.Errorf("simulate_claude_max_enabled only supported for anthropic groups")
+		}
+		group.SimulateClaudeMaxEnabled = *input.SimulateClaudeMaxEnabled
+	}
+	if group.Platform != PlatformAnthropic {
+		group.SimulateClaudeMaxEnabled = false
+	}
 
 	// 支持的模型系列（仅 antigravity 平台使用）
 	if input.SupportedModelScopes != nil {

backend/internal/service/admin_service_bulk_update_test.go

@@ -43,6 +43,16 @@ func (s *accountRepoStubForBulkUpdate) BindGroups(_ context.Context, accountID i
 	return nil
 }
 
+func (s *accountRepoStubForBulkUpdate) ListByGroup(_ context.Context, groupID int64) ([]Account, error) {
+	if err, ok := s.listByGroupErr[groupID]; ok {
+		return nil, err
+	}
+	if rows, ok := s.listByGroupData[groupID]; ok {
+		return rows, nil
+	}
+	return nil, nil
+}
+
 func (s *accountRepoStubForBulkUpdate) GetByIDs(_ context.Context, ids []int64) ([]*Account, error) {
 	s.getByIDsCalled = true
 	s.getByIDsIDs = append([]int64{}, ids...)
@@ -63,16 +73,6 @@ func (s *accountRepoStubForBulkUpdate) GetByID(_ context.Context, id int64) (*Ac
 	return nil, errors.New("account not found")
 }
 
-func (s *accountRepoStubForBulkUpdate) ListByGroup(_ context.Context, groupID int64) ([]Account, error) {
-	if err, ok := s.listByGroupErr[groupID]; ok {
-		return nil, err
-	}
-	if rows, ok := s.listByGroupData[groupID]; ok {
-		return rows, nil
-	}
-	return nil, nil
-}
-
 // TestAdminService_BulkUpdateAccounts_AllSuccessIDs 验证批量更新成功时返回 success_ids/failed_ids。
 func TestAdminService_BulkUpdateAccounts_AllSuccessIDs(t *testing.T) {
 	repo := &accountRepoStubForBulkUpdate{}

backend/internal/service/admin_service_group_test.go

@@ -785,3 +785,57 @@ func TestAdminService_UpdateGroup_InvalidRequestFallbackAllowsAntigravity(t *tes
 	require.NotNil(t, repo.updated)
 	require.Equal(t, fallbackID, *repo.updated.FallbackGroupIDOnInvalidRequest)
 }
+
+func TestAdminService_CreateGroup_SimulateClaudeMaxRequiresAnthropic(t *testing.T) {
+	repo := &groupRepoStubForAdmin{}
+	svc := &adminServiceImpl{groupRepo: repo}
+
+	enabled := true
+	_, err := svc.CreateGroup(context.Background(), &CreateGroupInput{
+		Name:                     "openai-group",
+		Platform:                 PlatformOpenAI,
+		SimulateClaudeMaxEnabled: &enabled,
+	})
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "simulate_claude_max_enabled only supported for anthropic groups")
+	require.Nil(t, repo.created)
+}
+
+func TestAdminService_UpdateGroup_SimulateClaudeMaxRequiresAnthropic(t *testing.T) {
+	existingGroup := &Group{
+		ID:       1,
+		Name:     "openai-group",
+		Platform: PlatformOpenAI,
+		Status:   StatusActive,
+	}
+	repo := &groupRepoStubForAdmin{getByID: existingGroup}
+	svc := &adminServiceImpl{groupRepo: repo}
+
+	enabled := true
+	_, err := svc.UpdateGroup(context.Background(), 1, &UpdateGroupInput{
+		SimulateClaudeMaxEnabled: &enabled,
+	})
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "simulate_claude_max_enabled only supported for anthropic groups")
+	require.Nil(t, repo.updated)
+}
+
+func TestAdminService_UpdateGroup_ClearsSimulateClaudeMaxWhenPlatformChanges(t *testing.T) {
+	existingGroup := &Group{
+		ID:                       1,
+		Name:                     "anthropic-group",
+		Platform:                 PlatformAnthropic,
+		Status:                   StatusActive,
+		SimulateClaudeMaxEnabled: true,
+	}
+	repo := &groupRepoStubForAdmin{getByID: existingGroup}
+	svc := &adminServiceImpl{groupRepo: repo}
+
+	group, err := svc.UpdateGroup(context.Background(), 1, &UpdateGroupInput{
+		Platform: PlatformOpenAI,
+	})
+	require.NoError(t, err)
+	require.NotNil(t, group)
+	require.NotNil(t, repo.updated)
+	require.False(t, repo.updated.SimulateClaudeMaxEnabled)
+}

backend/internal/service/antigravity_gateway_service.go

@@ -1599,7 +1599,7 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context,
 	var clientDisconnect bool
 	if claudeReq.Stream {
 		// 客户端要求流式，直接透传转换
-		streamRes, err := s.handleClaudeStreamingResponse(c, resp, startTime, originalModel)
+		streamRes, err := s.handleClaudeStreamingResponse(c, resp, startTime, originalModel, account.ID)
 		if err != nil {
 			logger.LegacyPrintf("service.antigravity_gateway", "%s status=stream_error error=%v", prefix, err)
 			return nil, err
@@ -1609,7 +1609,7 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context,
 		clientDisconnect = streamRes.clientDisconnect
 	} else {
 		// 客户端要求非流式，收集流式响应后转换返回
-		streamRes, err := s.handleClaudeStreamToNonStreaming(c, resp, startTime, originalModel)
+		streamRes, err := s.handleClaudeStreamToNonStreaming(c, resp, startTime, originalModel, account.ID)
 		if err != nil {
 			logger.LegacyPrintf("service.antigravity_gateway", "%s status=stream_collect_error error=%v", prefix, err)
 			return nil, err
@@ -1618,6 +1618,9 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context,
 		firstTokenMs = streamRes.firstTokenMs
 	}
 
+	// Claude Max cache billing: 同步 ForwardResult.Usage 与客户端响应体一致
+	applyClaudeMaxCacheBillingPolicyToUsage(usage, parsedRequestFromGinContext(c), claudeMaxGroupFromGinContext(c), originalModel, account.ID)
+
 	return &ForwardResult{
 		RequestID:        requestID,
 		Usage:            *usage,
@@ -3415,7 +3418,7 @@ func (s *AntigravityGatewayService) writeGoogleError(c *gin.Context, status int,
 
 // handleClaudeStreamToNonStreaming 收集上游流式响应，转换为 Claude 非流式格式返回
 // 用于处理客户端非流式请求但上游只支持流式的情况
-func (s *AntigravityGatewayService) handleClaudeStreamToNonStreaming(c *gin.Context, resp *http.Response, startTime time.Time, originalModel string) (*antigravityStreamResult, error) {
+func (s *AntigravityGatewayService) handleClaudeStreamToNonStreaming(c *gin.Context, resp *http.Response, startTime time.Time, originalModel string, accountID int64) (*antigravityStreamResult, error) {
 	scanner := bufio.NewScanner(resp.Body)
 	maxLineSize := defaultMaxLineSize
 	if s.settingService.cfg != nil && s.settingService.cfg.Gateway.MaxLineSize > 0 {
@@ -3573,6 +3576,9 @@ returnResponse:
 		return nil, s.writeClaudeError(c, http.StatusBadGateway, "upstream_error", "Failed to parse upstream response")
 	}
 
+	// Claude Max cache billing simulation (non-streaming)
+	claudeResp = applyClaudeMaxNonStreamingRewrite(c, claudeResp, agUsage, originalModel, accountID)
+
 	c.Data(http.StatusOK, "application/json", claudeResp)
 
 	// 转换为 service.ClaudeUsage
@@ -3587,7 +3593,7 @@ returnResponse:
 }
 
 // handleClaudeStreamingResponse 处理 Claude 流式响应（Gemini SSE → Claude SSE 转换）
-func (s *AntigravityGatewayService) handleClaudeStreamingResponse(c *gin.Context, resp *http.Response, startTime time.Time, originalModel string) (*antigravityStreamResult, error) {
+func (s *AntigravityGatewayService) handleClaudeStreamingResponse(c *gin.Context, resp *http.Response, startTime time.Time, originalModel string, accountID int64) (*antigravityStreamResult, error) {
 	c.Header("Content-Type", "text/event-stream")
 	c.Header("Cache-Control", "no-cache")
 	c.Header("Connection", "keep-alive")
@@ -3600,6 +3606,8 @@ func (s *AntigravityGatewayService) handleClaudeStreamingResponse(c *gin.Context
 	}
 
 	processor := antigravity.NewStreamingProcessor(originalModel)
+	setupClaudeMaxStreamingHook(c, processor, originalModel, accountID)
+
 	var firstTokenMs *int
 	// 使用 Scanner 并限制单行大小，避免 ReadString 无上限导致 OOM
 	scanner := bufio.NewScanner(resp.Body)

backend/internal/service/antigravity_gateway_service_test.go

@@ -710,7 +710,7 @@ func TestHandleClaudeStreamingResponse_NormalComplete(t *testing.T) {
 		fmt.Fprintln(pw, "")
 	}()
 
-	result, err := svc.handleClaudeStreamingResponse(c, resp, time.Now(), "claude-sonnet-4-5")
+	result, err := svc.handleClaudeStreamingResponse(c, resp, time.Now(), "claude-sonnet-4-5", 0)
 	_ = pr.Close()
 
 	require.NoError(t, err)
@@ -787,7 +787,7 @@ func TestHandleClaudeStreamingResponse_ThoughtsTokenCount(t *testing.T) {
 		fmt.Fprintln(pw, "")
 	}()
 
-	result, err := svc.handleClaudeStreamingResponse(c, resp, time.Now(), "gemini-2.5-pro")
+	result, err := svc.handleClaudeStreamingResponse(c, resp, time.Now(), "gemini-2.5-pro", 0)
 	_ = pr.Close()
 
 	require.NoError(t, err)
@@ -990,7 +990,7 @@ func TestHandleClaudeStreamingResponse_ClientDisconnect(t *testing.T) {
 		fmt.Fprintln(pw, "")
 	}()
 
-	result, err := svc.handleClaudeStreamingResponse(c, resp, time.Now(), "claude-sonnet-4-5")
+	result, err := svc.handleClaudeStreamingResponse(c, resp, time.Now(), "claude-sonnet-4-5", 0)
 	_ = pr.Close()
 
 	require.NoError(t, err)
@@ -1014,7 +1014,7 @@ func TestHandleClaudeStreamingResponse_ContextCanceled(t *testing.T) {
 
 	resp := &http.Response{StatusCode: http.StatusOK, Body: cancelReadCloser{}, Header: http.Header{}}
 
-	result, err := svc.handleClaudeStreamingResponse(c, resp, time.Now(), "claude-sonnet-4-5")
+	result, err := svc.handleClaudeStreamingResponse(c, resp, time.Now(), "claude-sonnet-4-5", 0)
 
 	require.NoError(t, err)
 	require.NotNil(t, result)

backend/internal/service/antigravity_smart_retry_test.go

@@ -9,6 +9,7 @@ import (
 	"net/http"
 	"strings"
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/require"
 )
@@ -29,6 +30,15 @@ func (c *stubSmartRetryCache) DeleteSessionAccountID(_ context.Context, groupID
 	c.deleteCalls = append(c.deleteCalls, deleteSessionCall{groupID: groupID, sessionHash: sessionHash})
 	return nil
 }
+func (c *stubSmartRetryCache) GetClientAffinityAccounts(_ context.Context, _ int64, _ string, _ time.Duration) ([]int64, error) {
+	return nil, nil
+}
+func (c *stubSmartRetryCache) UpdateClientAffinity(_ context.Context, _ int64, _ string, _ int64, _ time.Duration) error {
+	return nil
+}
+func (c *stubSmartRetryCache) RemoveClientAffinity(_ context.Context, _ int64, _ string, _ int64) error {
+	return nil
+}
 
 // mockSmartRetryUpstream 用于 handleSmartRetry 测试的 mock upstream
 type mockSmartRetryUpstream struct {

backend/internal/service/api_key_auth_cache.go

@@ -59,9 +59,10 @@ type APIKeyAuthGroupSnapshot struct {
 
 	// Model routing is used by gateway account selection, so it must be part of auth cache snapshot.
 	// Only anthropic groups use these fields; others may leave them empty.
-	ModelRouting        map[string][]int64 `json:"model_routing,omitempty"`
-	ModelRoutingEnabled bool               `json:"model_routing_enabled"`
-	MCPXMLInject        bool               `json:"mcp_xml_inject"`
+	ModelRouting             map[string][]int64 `json:"model_routing,omitempty"`
+	ModelRoutingEnabled      bool               `json:"model_routing_enabled"`
+	MCPXMLInject             bool               `json:"mcp_xml_inject"`
+	SimulateClaudeMaxEnabled bool               `json:"simulate_claude_max_enabled"`
 
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string `json:"supported_model_scopes,omitempty"`

backend/internal/service/api_key_auth_cache_impl.go

@@ -244,6 +244,7 @@ func (s *APIKeyService) snapshotFromAPIKey(apiKey *APIKey) *APIKeyAuthSnapshot {
 			ModelRouting:                    apiKey.Group.ModelRouting,
 			ModelRoutingEnabled:             apiKey.Group.ModelRoutingEnabled,
 			MCPXMLInject:                    apiKey.Group.MCPXMLInject,
+			SimulateClaudeMaxEnabled:        apiKey.Group.SimulateClaudeMaxEnabled,
 			SupportedModelScopes:            apiKey.Group.SupportedModelScopes,
 		}
 	}
@@ -301,6 +302,7 @@ func (s *APIKeyService) snapshotToAPIKey(key string, snapshot *APIKeyAuthSnapsho
 			ModelRouting:                    snapshot.Group.ModelRouting,
 			ModelRoutingEnabled:             snapshot.Group.ModelRoutingEnabled,
 			MCPXMLInject:                    snapshot.Group.MCPXMLInject,
+			SimulateClaudeMaxEnabled:        snapshot.Group.SimulateClaudeMaxEnabled,
 			SupportedModelScopes:            snapshot.Group.SupportedModelScopes,
 		}
 	}

backend/internal/service/claude_max_cache_billing_policy.go

@@ -0,0 +1,450 @@
+package service
+
+import (
+	"encoding/json"
+	"strings"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/claude"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	"github.com/tidwall/gjson"
+)
+
+type claudeMaxCacheBillingOutcome struct {
+	Simulated bool
+}
+
+func applyClaudeMaxCacheBillingPolicyToUsage(usage *ClaudeUsage, parsed *ParsedRequest, group *Group, model string, accountID int64) claudeMaxCacheBillingOutcome {
+	var out claudeMaxCacheBillingOutcome
+	if usage == nil || !shouldApplyClaudeMaxBillingRulesForUsage(group, model, parsed) {
+		return out
+	}
+
+	resolvedModel := strings.TrimSpace(model)
+	if resolvedModel == "" && parsed != nil {
+		resolvedModel = strings.TrimSpace(parsed.Model)
+	}
+
+	if hasCacheCreationTokens(*usage) {
+		// Upstream already returned cache creation usage; keep original usage.
+		return out
+	}
+
+	if !shouldSimulateClaudeMaxUsageForUsage(*usage, parsed) {
+		return out
+	}
+	beforeInputTokens := usage.InputTokens
+	out.Simulated = safelyProjectUsageToClaudeMax1H(usage, parsed)
+	if out.Simulated {
+		logger.LegacyPrintf("service.gateway", "simulate_claude_max_usage: model=%s account=%d input_tokens:%d->%d cache_creation_1h=%d",
+			resolvedModel,
+			accountID,
+			beforeInputTokens,
+			usage.InputTokens,
+			usage.CacheCreation1hTokens,
+		)
+	}
+	return out
+}
+
+func isClaudeFamilyModel(model string) bool {
+	normalized := strings.ToLower(strings.TrimSpace(claude.NormalizeModelID(model)))
+	if normalized == "" {
+		return false
+	}
+	return strings.Contains(normalized, "claude-")
+}
+
+func shouldApplyClaudeMaxBillingRules(input *RecordUsageInput) bool {
+	if input == nil || input.Result == nil || input.APIKey == nil || input.APIKey.Group == nil {
+		return false
+	}
+	return shouldApplyClaudeMaxBillingRulesForUsage(input.APIKey.Group, input.Result.Model, input.ParsedRequest)
+}
+
+func shouldApplyClaudeMaxBillingRulesForUsage(group *Group, model string, parsed *ParsedRequest) bool {
+	if group == nil {
+		return false
+	}
+	if !group.SimulateClaudeMaxEnabled || group.Platform != PlatformAnthropic {
+		return false
+	}
+
+	resolvedModel := model
+	if resolvedModel == "" && parsed != nil {
+		resolvedModel = parsed.Model
+	}
+	if !isClaudeFamilyModel(resolvedModel) {
+		return false
+	}
+	return true
+}
+
+func hasCacheCreationTokens(usage ClaudeUsage) bool {
+	return usage.CacheCreationInputTokens > 0 || usage.CacheCreation5mTokens > 0 || usage.CacheCreation1hTokens > 0
+}
+
+func shouldSimulateClaudeMaxUsage(input *RecordUsageInput) bool {
+	if input == nil || input.Result == nil {
+		return false
+	}
+	if !shouldApplyClaudeMaxBillingRules(input) {
+		return false
+	}
+	return shouldSimulateClaudeMaxUsageForUsage(input.Result.Usage, input.ParsedRequest)
+}
+
+func shouldSimulateClaudeMaxUsageForUsage(usage ClaudeUsage, parsed *ParsedRequest) bool {
+	if usage.InputTokens <= 0 {
+		return false
+	}
+	if hasCacheCreationTokens(usage) {
+		return false
+	}
+	if !hasClaudeCacheSignals(parsed) {
+		return false
+	}
+	return true
+}
+
+func safelyProjectUsageToClaudeMax1H(usage *ClaudeUsage, parsed *ParsedRequest) (changed bool) {
+	defer func() {
+		if r := recover(); r != nil {
+			logger.LegacyPrintf("service.gateway", "simulate_claude_max_usage skipped: panic=%v", r)
+			changed = false
+		}
+	}()
+	return projectUsageToClaudeMax1H(usage, parsed)
+}
+
+func projectUsageToClaudeMax1H(usage *ClaudeUsage, parsed *ParsedRequest) bool {
+	if usage == nil {
+		return false
+	}
+	totalWindowTokens := usage.InputTokens + usage.CacheCreation5mTokens + usage.CacheCreation1hTokens
+	if totalWindowTokens <= 1 {
+		return false
+	}
+
+	simulatedInputTokens := computeClaudeMaxProjectedInputTokens(totalWindowTokens, parsed)
+	if simulatedInputTokens <= 0 {
+		simulatedInputTokens = 1
+	}
+	if simulatedInputTokens >= totalWindowTokens {
+		simulatedInputTokens = totalWindowTokens - 1
+	}
+
+	cacheCreation1hTokens := totalWindowTokens - simulatedInputTokens
+	if usage.InputTokens == simulatedInputTokens &&
+		usage.CacheCreation5mTokens == 0 &&
+		usage.CacheCreation1hTokens == cacheCreation1hTokens &&
+		usage.CacheCreationInputTokens == cacheCreation1hTokens {
+		return false
+	}
+
+	usage.InputTokens = simulatedInputTokens
+	usage.CacheCreation5mTokens = 0
+	usage.CacheCreation1hTokens = cacheCreation1hTokens
+	usage.CacheCreationInputTokens = cacheCreation1hTokens
+	return true
+}
+
+type claudeCacheProjection struct {
+	HasBreakpoint        bool
+	BreakpointCount      int
+	TotalEstimatedTokens int
+	TailEstimatedTokens  int
+}
+
+func computeClaudeMaxProjectedInputTokens(totalWindowTokens int, parsed *ParsedRequest) int {
+	if totalWindowTokens <= 1 {
+		return totalWindowTokens
+	}
+
+	projection := analyzeClaudeCacheProjection(parsed)
+	if !projection.HasBreakpoint || projection.TotalEstimatedTokens <= 0 || projection.TailEstimatedTokens <= 0 {
+		return totalWindowTokens
+	}
+
+	totalEstimate := int64(projection.TotalEstimatedTokens)
+	tailEstimate := int64(projection.TailEstimatedTokens)
+	if tailEstimate > totalEstimate {
+		tailEstimate = totalEstimate
+	}
+
+	scaled := (int64(totalWindowTokens)*tailEstimate + totalEstimate/2) / totalEstimate
+	if scaled <= 0 {
+		scaled = 1
+	}
+	if scaled >= int64(totalWindowTokens) {
+		scaled = int64(totalWindowTokens - 1)
+	}
+	return int(scaled)
+}
+
+func hasClaudeCacheSignals(parsed *ParsedRequest) bool {
+	if parsed == nil {
+		return false
+	}
+	if hasTopLevelEphemeralCacheControl(parsed) {
+		return true
+	}
+	return countExplicitCacheBreakpoints(parsed) > 0
+}
+
+func hasTopLevelEphemeralCacheControl(parsed *ParsedRequest) bool {
+	if parsed == nil || len(parsed.Body) == 0 {
+		return false
+	}
+	cacheType := strings.TrimSpace(gjson.GetBytes(parsed.Body, "cache_control.type").String())
+	return strings.EqualFold(cacheType, "ephemeral")
+}
+
+func analyzeClaudeCacheProjection(parsed *ParsedRequest) claudeCacheProjection {
+	var projection claudeCacheProjection
+	if parsed == nil {
+		return projection
+	}
+
+	total := 0
+	lastBreakpointAt := -1
+
+	switch system := parsed.System.(type) {
+	case string:
+		total += claudeMaxMessageOverheadTokens + estimateClaudeTextTokens(system)
+	case []any:
+		for _, raw := range system {
+			block, ok := raw.(map[string]any)
+			if !ok {
+				total += claudeMaxUnknownContentTokens
+				continue
+			}
+			total += estimateClaudeBlockTokens(block)
+			if hasEphemeralCacheControl(block) {
+				lastBreakpointAt = total
+				projection.BreakpointCount++
+				projection.HasBreakpoint = true
+			}
+		}
+	}
+
+	for _, rawMsg := range parsed.Messages {
+		total += claudeMaxMessageOverheadTokens
+		msg, ok := rawMsg.(map[string]any)
+		if !ok {
+			total += claudeMaxUnknownContentTokens
+			continue
+		}
+		content, exists := msg["content"]
+		if !exists {
+			continue
+		}
+		msgTokens, msgLastBreak, msgBreakCount := estimateClaudeContentTokens(content)
+		total += msgTokens
+		if msgBreakCount > 0 {
+			lastBreakpointAt = total - msgTokens + msgLastBreak
+			projection.BreakpointCount += msgBreakCount
+			projection.HasBreakpoint = true
+		}
+	}
+
+	if total <= 0 {
+		total = 1
+	}
+	projection.TotalEstimatedTokens = total
+
+	if projection.HasBreakpoint && lastBreakpointAt >= 0 {
+		tail := total - lastBreakpointAt
+		if tail <= 0 {
+			tail = 1
+		}
+		projection.TailEstimatedTokens = tail
+		return projection
+	}
+
+	if hasTopLevelEphemeralCacheControl(parsed) {
+		tail := estimateLastUserMessageTokens(parsed)
+		if tail <= 0 {
+			tail = 1
+		}
+		projection.HasBreakpoint = true
+		projection.BreakpointCount = 1
+		projection.TailEstimatedTokens = tail
+	}
+	return projection
+}
+
+func countExplicitCacheBreakpoints(parsed *ParsedRequest) int {
+	if parsed == nil {
+		return 0
+	}
+	total := 0
+	if system, ok := parsed.System.([]any); ok {
+		for _, raw := range system {
+			if block, ok := raw.(map[string]any); ok && hasEphemeralCacheControl(block) {
+				total++
+			}
+		}
+	}
+	for _, rawMsg := range parsed.Messages {
+		msg, ok := rawMsg.(map[string]any)
+		if !ok {
+			continue
+		}
+		content, ok := msg["content"].([]any)
+		if !ok {
+			continue
+		}
+		for _, raw := range content {
+			if block, ok := raw.(map[string]any); ok && hasEphemeralCacheControl(block) {
+				total++
+			}
+		}
+	}
+	return total
+}
+
+func hasEphemeralCacheControl(block map[string]any) bool {
+	if block == nil {
+		return false
+	}
+	raw, ok := block["cache_control"]
+	if !ok || raw == nil {
+		return false
+	}
+	switch cc := raw.(type) {
+	case map[string]any:
+		cacheType, _ := cc["type"].(string)
+		return strings.EqualFold(strings.TrimSpace(cacheType), "ephemeral")
+	case map[string]string:
+		return strings.EqualFold(strings.TrimSpace(cc["type"]), "ephemeral")
+	default:
+		return false
+	}
+}
+
+func estimateClaudeContentTokens(content any) (tokens int, lastBreakAt int, breakpointCount int) {
+	switch value := content.(type) {
+	case string:
+		return estimateClaudeTextTokens(value), -1, 0
+	case []any:
+		total := 0
+		lastBreak := -1
+		breaks := 0
+		for _, raw := range value {
+			block, ok := raw.(map[string]any)
+			if !ok {
+				total += claudeMaxUnknownContentTokens
+				continue
+			}
+			total += estimateClaudeBlockTokens(block)
+			if hasEphemeralCacheControl(block) {
+				lastBreak = total
+				breaks++
+			}
+		}
+		return total, lastBreak, breaks
+	default:
+		return estimateStructuredTokens(value), -1, 0
+	}
+}
+
+func estimateClaudeBlockTokens(block map[string]any) int {
+	if block == nil {
+		return claudeMaxUnknownContentTokens
+	}
+	tokens := claudeMaxBlockOverheadTokens
+	blockType, _ := block["type"].(string)
+	switch blockType {
+	case "text":
+		if text, ok := block["text"].(string); ok {
+			tokens += estimateClaudeTextTokens(text)
+		}
+	case "tool_result":
+		if content, ok := block["content"]; ok {
+			nested, _, _ := estimateClaudeContentTokens(content)
+			tokens += nested
+		}
+	case "tool_use":
+		if name, ok := block["name"].(string); ok {
+			tokens += estimateClaudeTextTokens(name)
+		}
+		if input, ok := block["input"]; ok {
+			tokens += estimateStructuredTokens(input)
+		}
+	default:
+		if text, ok := block["text"].(string); ok {
+			tokens += estimateClaudeTextTokens(text)
+		} else if content, ok := block["content"]; ok {
+			nested, _, _ := estimateClaudeContentTokens(content)
+			tokens += nested
+		}
+	}
+	if tokens <= claudeMaxBlockOverheadTokens {
+		tokens += claudeMaxUnknownContentTokens
+	}
+	return tokens
+}
+
+func estimateLastUserMessageTokens(parsed *ParsedRequest) int {
+	if parsed == nil || len(parsed.Messages) == 0 {
+		return 0
+	}
+	for i := len(parsed.Messages) - 1; i >= 0; i-- {
+		msg, ok := parsed.Messages[i].(map[string]any)
+		if !ok {
+			continue
+		}
+		role, _ := msg["role"].(string)
+		if !strings.EqualFold(strings.TrimSpace(role), "user") {
+			continue
+		}
+		tokens, _, _ := estimateClaudeContentTokens(msg["content"])
+		return claudeMaxMessageOverheadTokens + tokens
+	}
+	return 0
+}
+
+func estimateStructuredTokens(v any) int {
+	if v == nil {
+		return 0
+	}
+	raw, err := json.Marshal(v)
+	if err != nil {
+		return claudeMaxUnknownContentTokens
+	}
+	return estimateClaudeTextTokens(string(raw))
+}
+
+func estimateClaudeTextTokens(text string) int {
+	if tokens, ok := estimateTokensByThirdPartyTokenizer(text); ok {
+		return tokens
+	}
+	return estimateClaudeTextTokensHeuristic(text)
+}
+
+func estimateClaudeTextTokensHeuristic(text string) int {
+	normalized := strings.Join(strings.Fields(strings.TrimSpace(text)), " ")
+	if normalized == "" {
+		return 0
+	}
+	asciiChars := 0
+	nonASCIIChars := 0
+	for _, r := range normalized {
+		if r <= 127 {
+			asciiChars++
+		} else {
+			nonASCIIChars++
+		}
+	}
+	tokens := nonASCIIChars
+	if asciiChars > 0 {
+		tokens += (asciiChars + 3) / 4
+	}
+	if words := len(strings.Fields(normalized)); words > tokens {
+		tokens = words
+	}
+	if tokens <= 0 {
+		return 1
+	}
+	return tokens
+}

backend/internal/service/claude_max_simulation_test.go

@@ -0,0 +1,156 @@
+package service
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestProjectUsageToClaudeMax1H_Conservation(t *testing.T) {
+	usage := &ClaudeUsage{
+		InputTokens:              1200,
+		CacheCreationInputTokens: 0,
+		CacheCreation5mTokens:    0,
+		CacheCreation1hTokens:    0,
+	}
+	parsed := &ParsedRequest{
+		Model: "claude-sonnet-4-5",
+		Messages: []any{
+			map[string]any{
+				"role": "user",
+				"content": []any{
+					map[string]any{
+						"type":          "text",
+						"text":          strings.Repeat("cached context ", 200),
+						"cache_control": map[string]any{"type": "ephemeral"},
+					},
+					map[string]any{
+						"type": "text",
+						"text": "summarize quickly",
+					},
+				},
+			},
+		},
+	}
+
+	changed := projectUsageToClaudeMax1H(usage, parsed)
+	if !changed {
+		t.Fatalf("expected usage to be projected")
+	}
+
+	total := usage.InputTokens + usage.CacheCreation5mTokens + usage.CacheCreation1hTokens
+	if total != 1200 {
+		t.Fatalf("total tokens changed: got=%d want=%d", total, 1200)
+	}
+	if usage.CacheCreation5mTokens != 0 {
+		t.Fatalf("cache_creation_5m should be 0, got=%d", usage.CacheCreation5mTokens)
+	}
+	if usage.InputTokens <= 0 || usage.InputTokens >= 1200 {
+		t.Fatalf("simulated input out of range, got=%d", usage.InputTokens)
+	}
+	if usage.InputTokens > 100 {
+		t.Fatalf("simulated input should stay near cache breakpoint tail, got=%d", usage.InputTokens)
+	}
+	if usage.CacheCreation1hTokens <= 0 {
+		t.Fatalf("cache_creation_1h should be > 0, got=%d", usage.CacheCreation1hTokens)
+	}
+	if usage.CacheCreationInputTokens != usage.CacheCreation1hTokens {
+		t.Fatalf("cache_creation_input_tokens mismatch: got=%d want=%d", usage.CacheCreationInputTokens, usage.CacheCreation1hTokens)
+	}
+}
+
+func TestComputeClaudeMaxProjectedInputTokens_Deterministic(t *testing.T) {
+	parsed := &ParsedRequest{
+		Model: "claude-opus-4-5",
+		Messages: []any{
+			map[string]any{
+				"role": "user",
+				"content": []any{
+					map[string]any{
+						"type":          "text",
+						"text":          "build context",
+						"cache_control": map[string]any{"type": "ephemeral"},
+					},
+					map[string]any{
+						"type": "text",
+						"text": "what is failing now",
+					},
+				},
+			},
+		},
+	}
+
+	got1 := computeClaudeMaxProjectedInputTokens(4096, parsed)
+	got2 := computeClaudeMaxProjectedInputTokens(4096, parsed)
+	if got1 != got2 {
+		t.Fatalf("non-deterministic input tokens: %d != %d", got1, got2)
+	}
+}
+
+func TestShouldSimulateClaudeMaxUsage(t *testing.T) {
+	group := &Group{
+		Platform:                 PlatformAnthropic,
+		SimulateClaudeMaxEnabled: true,
+	}
+	input := &RecordUsageInput{
+		Result: &ForwardResult{
+			Model: "claude-sonnet-4-5",
+			Usage: ClaudeUsage{
+				InputTokens:              3000,
+				CacheCreationInputTokens: 0,
+				CacheCreation5mTokens:    0,
+				CacheCreation1hTokens:    0,
+			},
+		},
+		ParsedRequest: &ParsedRequest{
+			Messages: []any{
+				map[string]any{
+					"role": "user",
+					"content": []any{
+						map[string]any{
+							"type":          "text",
+							"text":          "cached",
+							"cache_control": map[string]any{"type": "ephemeral"},
+						},
+						map[string]any{
+							"type": "text",
+							"text": "tail",
+						},
+					},
+				},
+			},
+		},
+		APIKey: &APIKey{Group: group},
+	}
+
+	if !shouldSimulateClaudeMaxUsage(input) {
+		t.Fatalf("expected simulate=true for claude group with cache signal")
+	}
+
+	input.ParsedRequest = &ParsedRequest{
+		Messages: []any{
+			map[string]any{"role": "user", "content": "no cache signal"},
+		},
+	}
+	if shouldSimulateClaudeMaxUsage(input) {
+		t.Fatalf("expected simulate=false when request has no cache signal")
+	}
+
+	input.ParsedRequest = &ParsedRequest{
+		Messages: []any{
+			map[string]any{
+				"role": "user",
+				"content": []any{
+					map[string]any{
+						"type":          "text",
+						"text":          "cached",
+						"cache_control": map[string]any{"type": "ephemeral"},
+					},
+				},
+			},
+		},
+	}
+	input.Result.Usage.CacheCreationInputTokens = 100
+	if shouldSimulateClaudeMaxUsage(input) {
+		t.Fatalf("expected simulate=false when cache creation already exists")
+	}
+}

backend/internal/service/claude_tokenizer.go

@@ -0,0 +1,41 @@
+package service
+
+import (
+	"sync"
+
+	tiktoken "github.com/pkoukk/tiktoken-go"
+	tiktokenloader "github.com/pkoukk/tiktoken-go-loader"
+)
+
+var (
+	claudeTokenizerOnce sync.Once
+	claudeTokenizer     *tiktoken.Tiktoken
+)
+
+func getClaudeTokenizer() *tiktoken.Tiktoken {
+	claudeTokenizerOnce.Do(func() {
+		// Use offline loader to avoid runtime dictionary download.
+		tiktoken.SetBpeLoader(tiktokenloader.NewOfflineLoader())
+		// Use a high-capacity tokenizer as the default approximation for Claude payloads.
+		enc, err := tiktoken.GetEncoding(tiktoken.MODEL_O200K_BASE)
+		if err != nil {
+			enc, err = tiktoken.GetEncoding(tiktoken.MODEL_CL100K_BASE)
+		}
+		if err == nil {
+			claudeTokenizer = enc
+		}
+	})
+	return claudeTokenizer
+}
+
+func estimateTokensByThirdPartyTokenizer(text string) (int, bool) {
+	enc := getClaudeTokenizer()
+	if enc == nil {
+		return 0, false
+	}
+	tokens := len(enc.EncodeOrdinary(text))
+	if tokens <= 0 {
+		return 0, false
+	}
+	return tokens, true
+}

backend/internal/service/concurrency_service.go

@@ -331,8 +331,9 @@ func (s *ConcurrencyService) StartSlotCleanupWorker(accountRepo AccountRepositor
 	}()
 }
 
-// GetAccountConcurrencyBatch gets current concurrency counts for multiple accounts
-// Returns a map of accountID -> current concurrency count
+// GetAccountConcurrencyBatch gets current concurrency counts for multiple accounts.
+// Uses a detached context with timeout to prevent HTTP request cancellation from
+// causing the entire batch to fail (which would show all concurrency as 0).
 func (s *ConcurrencyService) GetAccountConcurrencyBatch(ctx context.Context, accountIDs []int64) (map[int64]int, error) {
 	if len(accountIDs) == 0 {
 		return map[int64]int{}, nil
@@ -344,5 +345,11 @@ func (s *ConcurrencyService) GetAccountConcurrencyBatch(ctx context.Context, acc
 		}
 		return result, nil
 	}
-	return s.cache.GetAccountConcurrencyBatch(ctx, accountIDs)
+
+	// Use a detached context so that a cancelled HTTP request doesn't cause
+	// the Redis pipeline to fail and return all-zero concurrency counts.
+	redisCtx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
+	defer cancel()
+
+	return s.cache.GetAccountConcurrencyBatch(redisCtx, accountIDs)
 }

backend/internal/service/error_passthrough_runtime_test.go

@@ -220,7 +220,7 @@ func TestApplyErrorPassthroughRule_SkipMonitoringSetsContextKey(t *testing.T) {
 	v, exists := c.Get(OpsSkipPassthroughKey)
 	assert.True(t, exists, "OpsSkipPassthroughKey should be set when skip_monitoring=true")
 	boolVal, ok := v.(bool)
-	assert.True(t, ok, "value should be bool")
+	assert.True(t, ok, "value should be a bool")
 	assert.True(t, boolVal)
 }
 

backend/internal/service/gateway_claude_max_response_helpers.go

@@ -0,0 +1,196 @@
+package service
+
+import (
+	"context"
+	"encoding/json"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/antigravity"
+	"github.com/gin-gonic/gin"
+	"github.com/tidwall/sjson"
+)
+
+type claudeMaxResponseRewriteContext struct {
+	Parsed *ParsedRequest
+	Group  *Group
+}
+
+type claudeMaxResponseRewriteContextKeyType struct{}
+
+var claudeMaxResponseRewriteContextKey = claudeMaxResponseRewriteContextKeyType{}
+
+func withClaudeMaxResponseRewriteContext(ctx context.Context, c *gin.Context, parsed *ParsedRequest) context.Context {
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	value := claudeMaxResponseRewriteContext{
+		Parsed: parsed,
+		Group:  claudeMaxGroupFromGinContext(c),
+	}
+	return context.WithValue(ctx, claudeMaxResponseRewriteContextKey, value)
+}
+
+func claudeMaxResponseRewriteContextFromContext(ctx context.Context) claudeMaxResponseRewriteContext {
+	if ctx == nil {
+		return claudeMaxResponseRewriteContext{}
+	}
+	value, _ := ctx.Value(claudeMaxResponseRewriteContextKey).(claudeMaxResponseRewriteContext)
+	return value
+}
+
+func claudeMaxGroupFromGinContext(c *gin.Context) *Group {
+	if c == nil {
+		return nil
+	}
+	raw, exists := c.Get("api_key")
+	if !exists {
+		return nil
+	}
+	apiKey, ok := raw.(*APIKey)
+	if !ok || apiKey == nil {
+		return nil
+	}
+	return apiKey.Group
+}
+
+func parsedRequestFromGinContext(c *gin.Context) *ParsedRequest {
+	if c == nil {
+		return nil
+	}
+	raw, exists := c.Get("parsed_request")
+	if !exists {
+		return nil
+	}
+	parsed, _ := raw.(*ParsedRequest)
+	return parsed
+}
+
+func applyClaudeMaxSimulationToUsage(ctx context.Context, usage *ClaudeUsage, model string, accountID int64) claudeMaxCacheBillingOutcome {
+	var out claudeMaxCacheBillingOutcome
+	if usage == nil {
+		return out
+	}
+	rewriteCtx := claudeMaxResponseRewriteContextFromContext(ctx)
+	return applyClaudeMaxCacheBillingPolicyToUsage(usage, rewriteCtx.Parsed, rewriteCtx.Group, model, accountID)
+}
+
+func applyClaudeMaxSimulationToUsageJSONMap(ctx context.Context, usageObj map[string]any, model string, accountID int64) claudeMaxCacheBillingOutcome {
+	var out claudeMaxCacheBillingOutcome
+	if usageObj == nil {
+		return out
+	}
+	usage := claudeUsageFromJSONMap(usageObj)
+	out = applyClaudeMaxSimulationToUsage(ctx, &usage, model, accountID)
+	if out.Simulated {
+		rewriteClaudeUsageJSONMap(usageObj, usage)
+	}
+	return out
+}
+
+func rewriteClaudeUsageJSONBytes(body []byte, usage ClaudeUsage) []byte {
+	updated := body
+	var err error
+
+	updated, err = sjson.SetBytes(updated, "usage.input_tokens", usage.InputTokens)
+	if err != nil {
+		return body
+	}
+	updated, err = sjson.SetBytes(updated, "usage.cache_creation_input_tokens", usage.CacheCreationInputTokens)
+	if err != nil {
+		return body
+	}
+	updated, err = sjson.SetBytes(updated, "usage.cache_creation.ephemeral_5m_input_tokens", usage.CacheCreation5mTokens)
+	if err != nil {
+		return body
+	}
+	updated, err = sjson.SetBytes(updated, "usage.cache_creation.ephemeral_1h_input_tokens", usage.CacheCreation1hTokens)
+	if err != nil {
+		return body
+	}
+	return updated
+}
+
+func claudeUsageFromJSONMap(usageObj map[string]any) ClaudeUsage {
+	var usage ClaudeUsage
+	if usageObj == nil {
+		return usage
+	}
+
+	usage.InputTokens = usageIntFromAny(usageObj["input_tokens"])
+	usage.OutputTokens = usageIntFromAny(usageObj["output_tokens"])
+	usage.CacheCreationInputTokens = usageIntFromAny(usageObj["cache_creation_input_tokens"])
+	usage.CacheReadInputTokens = usageIntFromAny(usageObj["cache_read_input_tokens"])
+
+	if ccObj, ok := usageObj["cache_creation"].(map[string]any); ok {
+		usage.CacheCreation5mTokens = usageIntFromAny(ccObj["ephemeral_5m_input_tokens"])
+		usage.CacheCreation1hTokens = usageIntFromAny(ccObj["ephemeral_1h_input_tokens"])
+	}
+	return usage
+}
+
+func rewriteClaudeUsageJSONMap(usageObj map[string]any, usage ClaudeUsage) {
+	if usageObj == nil {
+		return
+	}
+	usageObj["input_tokens"] = usage.InputTokens
+	usageObj["cache_creation_input_tokens"] = usage.CacheCreationInputTokens
+
+	ccObj, _ := usageObj["cache_creation"].(map[string]any)
+	if ccObj == nil {
+		ccObj = make(map[string]any, 2)
+		usageObj["cache_creation"] = ccObj
+	}
+	ccObj["ephemeral_5m_input_tokens"] = usage.CacheCreation5mTokens
+	ccObj["ephemeral_1h_input_tokens"] = usage.CacheCreation1hTokens
+}
+
+func usageIntFromAny(v any) int {
+	switch value := v.(type) {
+	case int:
+		return value
+	case int64:
+		return int(value)
+	case float64:
+		return int(value)
+	case json.Number:
+		if n, err := value.Int64(); err == nil {
+			return int(n)
+		}
+	}
+	return 0
+}
+
+// setupClaudeMaxStreamingHook 为 Antigravity 流式路径设置 SSE usage 改写 hook。
+func setupClaudeMaxStreamingHook(c *gin.Context, processor *antigravity.StreamingProcessor, originalModel string, accountID int64) {
+	group := claudeMaxGroupFromGinContext(c)
+	parsed := parsedRequestFromGinContext(c)
+	if !shouldApplyClaudeMaxBillingRulesForUsage(group, originalModel, parsed) {
+		return
+	}
+	processor.SetUsageMapHook(func(usageMap map[string]any) {
+		svcUsage := claudeUsageFromJSONMap(usageMap)
+		outcome := applyClaudeMaxCacheBillingPolicyToUsage(&svcUsage, parsed, group, originalModel, accountID)
+		if outcome.Simulated {
+			rewriteClaudeUsageJSONMap(usageMap, svcUsage)
+		}
+	})
+}
+
+// applyClaudeMaxNonStreamingRewrite 为 Antigravity 非流式路径改写响应体中的 usage。
+func applyClaudeMaxNonStreamingRewrite(c *gin.Context, claudeResp []byte, agUsage *antigravity.ClaudeUsage, originalModel string, accountID int64) []byte {
+	group := claudeMaxGroupFromGinContext(c)
+	parsed := parsedRequestFromGinContext(c)
+	if !shouldApplyClaudeMaxBillingRulesForUsage(group, originalModel, parsed) {
+		return claudeResp
+	}
+	svcUsage := &ClaudeUsage{
+		InputTokens:              agUsage.InputTokens,
+		OutputTokens:             agUsage.OutputTokens,
+		CacheCreationInputTokens: agUsage.CacheCreationInputTokens,
+		CacheReadInputTokens:     agUsage.CacheReadInputTokens,
+	}
+	outcome := applyClaudeMaxCacheBillingPolicyToUsage(svcUsage, parsed, group, originalModel, accountID)
+	if outcome.Simulated {
+		return rewriteClaudeUsageJSONBytes(claudeResp, *svcUsage)
+	}
+	return claudeResp
+}

backend/internal/service/gateway_hotpath_optimization_test.go

@@ -143,6 +143,15 @@ func (s *stickyGatewayCacheHotpathStub) RefreshSessionTTL(ctx context.Context, g
 func (s *stickyGatewayCacheHotpathStub) DeleteSessionAccountID(ctx context.Context, groupID int64, sessionHash string) error {
 	return nil
 }
+func (s *stickyGatewayCacheHotpathStub) GetClientAffinityAccounts(_ context.Context, _ int64, _ string, _ time.Duration) ([]int64, error) {
+	return nil, nil
+}
+func (s *stickyGatewayCacheHotpathStub) UpdateClientAffinity(_ context.Context, _ int64, _ string, _ int64, _ time.Duration) error {
+	return nil
+}
+func (s *stickyGatewayCacheHotpathStub) RemoveClientAffinity(_ context.Context, _ int64, _ string, _ int64) error {
+	return nil
+}
 
 func (s *modelsListAccountRepoStub) ListSchedulableByGroupID(ctx context.Context, groupID int64) ([]Account, error) {
 	s.listByGroupCalls.Add(1)

backend/internal/service/gateway_multiplatform_test.go

@@ -235,6 +235,16 @@ func (m *mockGatewayCacheForPlatform) DeleteSessionAccountID(ctx context.Context
 	return nil
 }
 
+func (m *mockGatewayCacheForPlatform) GetClientAffinityAccounts(_ context.Context, _ int64, _ string, _ time.Duration) ([]int64, error) {
+	return nil, nil
+}
+func (m *mockGatewayCacheForPlatform) UpdateClientAffinity(_ context.Context, _ int64, _ string, _ int64, _ time.Duration) error {
+	return nil
+}
+func (m *mockGatewayCacheForPlatform) RemoveClientAffinity(_ context.Context, _ int64, _ string, _ int64) error {
+	return nil
+}
+
 type mockGroupRepoForGateway struct {
 	groups           map[int64]*Group
 	getByIDCalls     int

backend/internal/service/gateway_record_usage_claude_max_test.go

@@ -0,0 +1,199 @@
+package service
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/stretchr/testify/require"
+)
+
+type usageLogRepoRecordUsageStub struct {
+	UsageLogRepository
+
+	last     *UsageLog
+	inserted bool
+	err      error
+}
+
+func (s *usageLogRepoRecordUsageStub) Create(_ context.Context, log *UsageLog) (bool, error) {
+	copied := *log
+	s.last = &copied
+	return s.inserted, s.err
+}
+
+func newGatewayServiceForRecordUsageTest(repo UsageLogRepository) *GatewayService {
+	return &GatewayService{
+		usageLogRepo:    repo,
+		billingService:  NewBillingService(&config.Config{}, nil),
+		cfg:             &config.Config{RunMode: config.RunModeSimple},
+		deferredService: &DeferredService{},
+	}
+}
+
+func TestRecordUsage_SimulateClaudeMaxEnabled_ProjectsUsageAndSkipsTTLOverride(t *testing.T) {
+	repo := &usageLogRepoRecordUsageStub{inserted: true}
+	svc := newGatewayServiceForRecordUsageTest(repo)
+
+	groupID := int64(11)
+	input := &RecordUsageInput{
+		Result: &ForwardResult{
+			RequestID: "req-sim-1",
+			Model:     "claude-sonnet-4",
+			Duration:  time.Second,
+			Usage: ClaudeUsage{
+				InputTokens: 160,
+			},
+		},
+		ParsedRequest: &ParsedRequest{
+			Model: "claude-sonnet-4",
+			Messages: []any{
+				map[string]any{
+					"role": "user",
+					"content": []any{
+						map[string]any{
+							"type":          "text",
+							"text":          "long cached context for prior turns",
+							"cache_control": map[string]any{"type": "ephemeral"},
+						},
+						map[string]any{
+							"type": "text",
+							"text": "please summarize the logs and provide root cause analysis",
+						},
+					},
+				},
+			},
+		},
+		APIKey: &APIKey{
+			ID:      1,
+			GroupID: &groupID,
+			Group: &Group{
+				ID:                       groupID,
+				Platform:                 PlatformAnthropic,
+				RateMultiplier:           1,
+				SimulateClaudeMaxEnabled: true,
+			},
+		},
+		User: &User{ID: 2},
+		Account: &Account{
+			ID:       3,
+			Platform: PlatformAnthropic,
+			Type:     AccountTypeOAuth,
+			Extra: map[string]any{
+				"cache_ttl_override_enabled": true,
+				"cache_ttl_override_target":  "5m",
+			},
+		},
+	}
+
+	err := svc.RecordUsage(context.Background(), input)
+	require.NoError(t, err)
+	require.NotNil(t, repo.last)
+
+	log := repo.last
+	require.Equal(t, 80, log.InputTokens)
+	require.Equal(t, 80, log.CacheCreationTokens)
+	require.Equal(t, 0, log.CacheCreation5mTokens)
+	require.Equal(t, 80, log.CacheCreation1hTokens)
+	require.False(t, log.CacheTTLOverridden, "simulate outcome should skip account ttl override")
+}
+
+func TestRecordUsage_SimulateClaudeMaxDisabled_AppliesTTLOverride(t *testing.T) {
+	repo := &usageLogRepoRecordUsageStub{inserted: true}
+	svc := newGatewayServiceForRecordUsageTest(repo)
+
+	groupID := int64(12)
+	input := &RecordUsageInput{
+		Result: &ForwardResult{
+			RequestID: "req-sim-2",
+			Model:     "claude-sonnet-4",
+			Duration:  time.Second,
+			Usage: ClaudeUsage{
+				InputTokens:              40,
+				CacheCreationInputTokens: 120,
+				CacheCreation1hTokens:    120,
+			},
+		},
+		APIKey: &APIKey{
+			ID:      2,
+			GroupID: &groupID,
+			Group: &Group{
+				ID:                       groupID,
+				Platform:                 PlatformAnthropic,
+				RateMultiplier:           1,
+				SimulateClaudeMaxEnabled: false,
+			},
+		},
+		User: &User{ID: 3},
+		Account: &Account{
+			ID:       4,
+			Platform: PlatformAnthropic,
+			Type:     AccountTypeOAuth,
+			Extra: map[string]any{
+				"cache_ttl_override_enabled": true,
+				"cache_ttl_override_target":  "5m",
+			},
+		},
+	}
+
+	err := svc.RecordUsage(context.Background(), input)
+	require.NoError(t, err)
+	require.NotNil(t, repo.last)
+
+	log := repo.last
+	require.Equal(t, 120, log.CacheCreationTokens)
+	require.Equal(t, 120, log.CacheCreation5mTokens)
+	require.Equal(t, 0, log.CacheCreation1hTokens)
+	require.True(t, log.CacheTTLOverridden)
+}
+
+func TestRecordUsage_SimulateClaudeMaxEnabled_ExistingCacheCreationBypassesSimulation(t *testing.T) {
+	repo := &usageLogRepoRecordUsageStub{inserted: true}
+	svc := newGatewayServiceForRecordUsageTest(repo)
+
+	groupID := int64(13)
+	input := &RecordUsageInput{
+		Result: &ForwardResult{
+			RequestID: "req-sim-3",
+			Model:     "claude-sonnet-4",
+			Duration:  time.Second,
+			Usage: ClaudeUsage{
+				InputTokens:              20,
+				CacheCreationInputTokens: 120,
+				CacheCreation5mTokens:    120,
+			},
+		},
+		APIKey: &APIKey{
+			ID:      3,
+			GroupID: &groupID,
+			Group: &Group{
+				ID:                       groupID,
+				Platform:                 PlatformAnthropic,
+				RateMultiplier:           1,
+				SimulateClaudeMaxEnabled: true,
+			},
+		},
+		User: &User{ID: 4},
+		Account: &Account{
+			ID:       5,
+			Platform: PlatformAnthropic,
+			Type:     AccountTypeOAuth,
+			Extra: map[string]any{
+				"cache_ttl_override_enabled": true,
+				"cache_ttl_override_target":  "5m",
+			},
+		},
+	}
+
+	err := svc.RecordUsage(context.Background(), input)
+	require.NoError(t, err)
+	require.NotNil(t, repo.last)
+
+	log := repo.last
+	require.Equal(t, 20, log.InputTokens)
+	require.Equal(t, 120, log.CacheCreation5mTokens)
+	require.Equal(t, 0, log.CacheCreation1hTokens)
+	require.Equal(t, 120, log.CacheCreationTokens)
+	require.False(t, log.CacheTTLOverridden, "existing cache_creation with SimulateClaudeMax enabled should skip account ttl override")
+}

backend/internal/service/gateway_response_usage_sync_test.go

@@ -0,0 +1,170 @@
+package service
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+)
+
+func TestHandleNonStreamingResponse_UsageAlignedWithClaudeMaxSimulation(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	svc := &GatewayService{
+		cfg:              &config.Config{},
+		rateLimitService: &RateLimitService{},
+	}
+
+	account := &Account{
+		ID:       11,
+		Platform: PlatformAnthropic,
+		Type:     AccountTypeOAuth,
+		Extra: map[string]any{
+			"cache_ttl_override_enabled": true,
+			"cache_ttl_override_target":  "5m",
+		},
+	}
+	group := &Group{
+		ID:                       99,
+		Platform:                 PlatformAnthropic,
+		SimulateClaudeMaxEnabled: true,
+	}
+	parsed := &ParsedRequest{
+		Model: "claude-sonnet-4",
+		Messages: []any{
+			map[string]any{
+				"role": "user",
+				"content": []any{
+					map[string]any{
+						"type":          "text",
+						"text":          "long cached context",
+						"cache_control": map[string]any{"type": "ephemeral"},
+					},
+					map[string]any{
+						"type": "text",
+						"text": "new user question",
+					},
+				},
+			},
+		},
+	}
+
+	upstreamBody := []byte(`{"id":"msg_1","model":"claude-sonnet-4","usage":{"input_tokens":120,"output_tokens":8}}`)
+	resp := &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"application/json"}},
+		Body:       ioNopCloserBytes(upstreamBody),
+	}
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", bytes.NewReader(nil))
+	c.Set("api_key", &APIKey{Group: group})
+	requestCtx := withClaudeMaxResponseRewriteContext(context.Background(), c, parsed)
+
+	usage, err := svc.handleNonStreamingResponse(requestCtx, resp, c, account, "claude-sonnet-4", "claude-sonnet-4")
+	require.NoError(t, err)
+	require.NotNil(t, usage)
+
+	var rendered struct {
+		Usage ClaudeUsage `json:"usage"`
+	}
+	require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &rendered))
+	rendered.Usage.CacheCreation5mTokens = int(gjson.GetBytes(rec.Body.Bytes(), "usage.cache_creation.ephemeral_5m_input_tokens").Int())
+	rendered.Usage.CacheCreation1hTokens = int(gjson.GetBytes(rec.Body.Bytes(), "usage.cache_creation.ephemeral_1h_input_tokens").Int())
+
+	require.Equal(t, rendered.Usage.InputTokens, usage.InputTokens)
+	require.Equal(t, rendered.Usage.OutputTokens, usage.OutputTokens)
+	require.Equal(t, rendered.Usage.CacheCreationInputTokens, usage.CacheCreationInputTokens)
+	require.Equal(t, rendered.Usage.CacheCreation5mTokens, usage.CacheCreation5mTokens)
+	require.Equal(t, rendered.Usage.CacheCreation1hTokens, usage.CacheCreation1hTokens)
+	require.Equal(t, rendered.Usage.CacheReadInputTokens, usage.CacheReadInputTokens)
+
+	require.Greater(t, usage.CacheCreation1hTokens, 0)
+	require.Equal(t, 0, usage.CacheCreation5mTokens)
+	require.Less(t, usage.InputTokens, 120)
+}
+
+func TestHandleNonStreamingResponse_ClaudeMaxDisabled_NoSimulationIntercept(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	svc := &GatewayService{
+		cfg:              &config.Config{},
+		rateLimitService: &RateLimitService{},
+	}
+
+	account := &Account{
+		ID:       12,
+		Platform: PlatformAnthropic,
+		Type:     AccountTypeOAuth,
+		Extra: map[string]any{
+			"cache_ttl_override_enabled": true,
+			"cache_ttl_override_target":  "5m",
+		},
+	}
+	group := &Group{
+		ID:                       100,
+		Platform:                 PlatformAnthropic,
+		SimulateClaudeMaxEnabled: false,
+	}
+	parsed := &ParsedRequest{
+		Model: "claude-sonnet-4",
+		Messages: []any{
+			map[string]any{
+				"role": "user",
+				"content": []any{
+					map[string]any{
+						"type":          "text",
+						"text":          "long cached context",
+						"cache_control": map[string]any{"type": "ephemeral"},
+					},
+					map[string]any{
+						"type": "text",
+						"text": "new user question",
+					},
+				},
+			},
+		},
+	}
+
+	upstreamBody := []byte(`{"id":"msg_2","model":"claude-sonnet-4","usage":{"input_tokens":120,"output_tokens":8}}`)
+	resp := &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"application/json"}},
+		Body:       ioNopCloserBytes(upstreamBody),
+	}
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", bytes.NewReader(nil))
+	c.Set("api_key", &APIKey{Group: group})
+	requestCtx := withClaudeMaxResponseRewriteContext(context.Background(), c, parsed)
+
+	usage, err := svc.handleNonStreamingResponse(requestCtx, resp, c, account, "claude-sonnet-4", "claude-sonnet-4")
+	require.NoError(t, err)
+	require.NotNil(t, usage)
+
+	require.Equal(t, 120, usage.InputTokens)
+	require.Equal(t, 0, usage.CacheCreationInputTokens)
+	require.Equal(t, 0, usage.CacheCreation5mTokens)
+	require.Equal(t, 0, usage.CacheCreation1hTokens)
+}
+
+func ioNopCloserBytes(b []byte) *readCloserFromBytes {
+	return &readCloserFromBytes{Reader: bytes.NewReader(b)}
+}
+
+type readCloserFromBytes struct {
+	*bytes.Reader
+}
+
+func (r *readCloserFromBytes) Close() error {
+	return nil
+}

backend/internal/service/gateway_service.go

@@ -56,6 +56,12 @@ const (
 	claudeMimicDebugInfoKey = "claude_mimic_debug_info"
 )
 
+const (
+	claudeMaxMessageOverheadTokens = 3
+	claudeMaxBlockOverheadTokens   = 1
+	claudeMaxUnknownContentTokens  = 4
+)
+
 // ForceCacheBillingContextKey 强制缓存计费上下文键
 // 用于粘性会话切换时，将 input_tokens 转为 cache_read_input_tokens 计费
 type forceCacheBillingKeyType struct{}
@@ -298,6 +304,10 @@ var (
 	sessionIDRegex       = regexp.MustCompile(`session_([a-f0-9-]{36})`)
 	claudeCliUserAgentRe = regexp.MustCompile(`^claude-cli/\d+\.\d+\.\d+`)
 
+	// clientIDFromMetadataRegex 从 metadata.user_id 中提取客户端 ID（64位 hex）
+	// 格式: user_{64位hex}_account_...
+	clientIDFromMetadataRegex = regexp.MustCompile(`^user_([a-f0-9]{64})_account_`)
+
 	// claudeCodePromptPrefixes 用于检测 Claude Code 系统提示词的前缀列表
 	// 支持多种变体：标准版、Agent SDK 版、Explore Agent 版、Compact 版等
 	// 注意：前缀之间不应存在包含关系，否则会导致冗余匹配
@@ -358,6 +368,13 @@ type GatewayCache interface {
 	// DeleteSessionAccountID 删除粘性会话绑定，用于账号不可用时主动清理
 	// Delete sticky session binding, used to proactively clean up when account becomes unavailable
 	DeleteSessionAccountID(ctx context.Context, groupID int64, sessionHash string) error
+
+	// GetClientAffinityAccounts 获取客户端亲和账号列表（按最近使用降序），同时清理过期成员
+	GetClientAffinityAccounts(ctx context.Context, groupID int64, clientID string, ttl time.Duration) ([]int64, error)
+	// UpdateClientAffinity 添加/更新客户端亲和关系（更新 score 为当前时间戳，刷新 key TTL）
+	UpdateClientAffinity(ctx context.Context, groupID int64, clientID string, accountID int64, ttl time.Duration) error
+	// RemoveClientAffinity 删除单个客户端亲和关系
+	RemoveClientAffinity(ctx context.Context, groupID int64, clientID string, accountID int64) error
 }
 
 // derefGroupID safely dereferences *int64 to int64, returning 0 if nil
@@ -428,6 +445,20 @@ func shouldClearStickySession(account *Account, requestedModel string) bool {
 	return false
 }
 
+// extractClientIDFromMetadata 从 metadata.user_id 中提取客户端 ID（64位 hex）。
+// 格式: user_{64位hex}_account_..._session_...
+// 返回空字符串表示无法提取（非 Claude Code/Console 客户端）。
+func extractClientIDFromMetadata(metadataUserID string) string {
+	if metadataUserID == "" {
+		return ""
+	}
+	matches := clientIDFromMetadataRegex.FindStringSubmatch(metadataUserID)
+	if matches == nil {
+		return ""
+	}
+	return matches[1]
+}
+
 type AccountWaitPlan struct {
 	AccountID      int64
 	MaxConcurrency int
@@ -1040,7 +1071,7 @@ func (s *GatewayService) SelectAccountForModelWithExclusions(ctx context.Context
 }
 
 // SelectAccountWithLoadAwareness selects account with load-awareness and wait plan.
-// metadataUserID: 已废弃参数，会话限制现在统一使用 sessionHash
+// metadataUserID: 用于客户端亲和调度，从中提取客户端 ID
 func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, groupID *int64, sessionHash string, requestedModel string, excludedIDs map[int64]struct{}, metadataUserID string) (*AccountSelectionResult, error) {
 	// 调试日志：记录调度入口参数
 	excludedIDsList := make([]int64, 0, len(excludedIDs))
@@ -1071,6 +1102,9 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 		}
 	}
 
+	// 提取客户端 ID（用于客户端亲和调度）
+	affinityClientID := extractClientIDFromMetadata(metadataUserID)
+
 	if s.debugModelRoutingEnabled() && requestedModel != "" {
 		groupPlatform := ""
 		if group != nil {
@@ -1368,6 +1402,9 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 						if sessionHash != "" && s.cache != nil {
 							_ = s.cache.SetSessionAccountID(ctx, derefGroupID(groupID), sessionHash, item.account.ID, stickySessionTTL)
 						}
+						if affinityClientID != "" && s.cache != nil && item.account.IsClientAffinityEnabled() {
+							_ = s.cache.UpdateClientAffinity(ctx, derefGroupID(groupID), affinityClientID, item.account.ID, stickySessionTTL)
+						}
 						if s.debugModelRoutingEnabled() {
 							logger.LegacyPrintf("service.gateway", "[ModelRoutingDebug] routed select: group_id=%v model=%s session=%s account=%d", derefGroupID(groupID), requestedModel, shortSessionHash(sessionHash), item.account.ID)
 						}
@@ -1464,6 +1501,66 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 		}
 	}
 
+	// ============ Layer 1.6: 客户端亲和（仅在粘性会话未命中时生效） ============
+	if affinityClientID != "" && s.cache != nil && stickyAccountID <= 0 {
+		affinityAccountIDs, err := s.cache.GetClientAffinityAccounts(ctx, derefGroupID(groupID), affinityClientID, stickySessionTTL)
+		if err == nil && len(affinityAccountIDs) > 0 {
+			for _, affinityAccID := range affinityAccountIDs {
+				if isExcluded(affinityAccID) {
+					continue
+				}
+				account, ok := accountByID[affinityAccID]
+				if !ok || !s.isAccountSchedulableForSelection(account) {
+					continue
+				}
+				if !account.IsClientAffinityEnabled() {
+					continue
+				}
+				if !s.isAccountAllowedForPlatform(account, platform, useMixed) {
+					continue
+				}
+				if requestedModel != "" && !s.isModelSupportedByAccountWithContext(ctx, account, requestedModel) {
+					continue
+				}
+				if !s.isAccountSchedulableForModelSelection(ctx, account, requestedModel) {
+					continue
+				}
+				if !s.isAccountSchedulableForQuota(account) {
+					continue
+				}
+				if !s.isAccountSchedulableForWindowCost(ctx, account, false) {
+					continue
+				}
+				if !s.isAccountSchedulableForRPM(ctx, account, false) {
+					continue
+				}
+
+				result, err := s.tryAcquireAccountSlot(ctx, affinityAccID, account.Concurrency)
+				if err == nil && result.Acquired {
+					if !s.checkAndRegisterSession(ctx, account, sessionHash) {
+						result.ReleaseFunc()
+						continue
+					}
+					// 亲和命中：更新亲和 score + 绑定粘性会话
+					_ = s.cache.UpdateClientAffinity(ctx, derefGroupID(groupID), affinityClientID, affinityAccID, stickySessionTTL)
+					if sessionHash != "" {
+						_ = s.cache.SetSessionAccountID(ctx, derefGroupID(groupID), sessionHash, affinityAccID, stickySessionTTL)
+					}
+					slog.Debug("client_affinity_hit",
+						"group_id", derefGroupID(groupID),
+						"client_id", affinityClientID[:8]+"...",
+						"account_id", affinityAccID)
+					return &AccountSelectionResult{
+						Account:     account,
+						Acquired:    true,
+						ReleaseFunc: result.ReleaseFunc,
+					}, nil
+				}
+			}
+			// 所有亲和账号不可用，继续到 Layer 2
+		}
+	}
+
 	// ============ Layer 2: 负载感知选择 ============
 	candidates := make([]*Account, 0, len(accounts))
 	for i := range accounts {
@@ -1516,6 +1613,9 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 	loadMap, err := s.concurrencyService.GetAccountsLoadBatch(ctx, accountLoads)
 	if err != nil {
 		if result, ok := s.tryAcquireByLegacyOrder(ctx, candidates, groupID, sessionHash, preferOAuth); ok {
+			if affinityClientID != "" && s.cache != nil && result.Account != nil && result.Account.IsClientAffinityEnabled() {
+				_ = s.cache.UpdateClientAffinity(ctx, derefGroupID(groupID), affinityClientID, result.Account.ID, stickySessionTTL)
+			}
 			return result, nil
 		}
 	} else {
@@ -1554,6 +1654,10 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 					if sessionHash != "" && s.cache != nil {
 						_ = s.cache.SetSessionAccountID(ctx, derefGroupID(groupID), sessionHash, selected.account.ID, stickySessionTTL)
 					}
+					// 更新客户端亲和关系
+					if affinityClientID != "" && s.cache != nil && selected.account.IsClientAffinityEnabled() {
+						_ = s.cache.UpdateClientAffinity(ctx, derefGroupID(groupID), affinityClientID, selected.account.ID, stickySessionTTL)
+					}
 					return &AccountSelectionResult{
 						Account:     selected.account,
 						Acquired:    true,
@@ -4348,6 +4452,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 	}
 
 	// 处理正常响应
+	ctx = withClaudeMaxResponseRewriteContext(ctx, c, parsed)
 
 	// 触发上游接受回调（提前释放串行锁，不等流完成）
 	if parsed.OnUpstreamAccepted != nil {
@@ -5804,6 +5909,7 @@ func (s *GatewayService) handleStreamingResponse(ctx context.Context, resp *http
 
 	needModelReplace := originalModel != mappedModel
 	clientDisconnected := false // 客户端断开标志，断开后继续读取上游以获取完整usage
+	skipAccountTTLOverride := false
 
 	pendingEventLines := make([]string, 0, 4)
 
@@ -5864,17 +5970,25 @@ func (s *GatewayService) handleStreamingResponse(ctx context.Context, resp *http
 			if msg, ok := event["message"].(map[string]any); ok {
 				if u, ok := msg["usage"].(map[string]any); ok {
 					eventChanged = reconcileCachedTokens(u) || eventChanged
+					claudeMaxOutcome := applyClaudeMaxSimulationToUsageJSONMap(ctx, u, originalModel, account.ID)
+					if claudeMaxOutcome.Simulated {
+						skipAccountTTLOverride = true
+					}
 				}
 			}
 		}
 		if eventType == "message_delta" {
 			if u, ok := event["usage"].(map[string]any); ok {
 				eventChanged = reconcileCachedTokens(u) || eventChanged
+				claudeMaxOutcome := applyClaudeMaxSimulationToUsageJSONMap(ctx, u, originalModel, account.ID)
+				if claudeMaxOutcome.Simulated {
+					skipAccountTTLOverride = true
+				}
 			}
 		}
 
 		// Cache TTL Override: 重写 SSE 事件中的 cache_creation 分类
-		if account.IsCacheTTLOverrideEnabled() {
+		if account.IsCacheTTLOverrideEnabled() && !skipAccountTTLOverride {
 			overrideTarget := account.GetCacheTTLOverrideTarget()
 			if eventType == "message_start" {
 				if msg, ok := event["message"].(map[string]any); ok {
@@ -6284,8 +6398,13 @@ func (s *GatewayService) handleNonStreamingResponse(ctx context.Context, resp *h
 		}
 	}
 
+	claudeMaxOutcome := applyClaudeMaxSimulationToUsage(ctx, &response.Usage, originalModel, account.ID)
+	if claudeMaxOutcome.Simulated {
+		body = rewriteClaudeUsageJSONBytes(body, response.Usage)
+	}
+
 	// Cache TTL Override: 重写 non-streaming 响应中的 cache_creation 分类
-	if account.IsCacheTTLOverrideEnabled() {
+	if account.IsCacheTTLOverrideEnabled() && !claudeMaxOutcome.Simulated {
 		overrideTarget := account.GetCacheTTLOverrideTarget()
 		if applyCacheTTLOverride(&response.Usage, overrideTarget) {
 			// 同步更新 body JSON 中的嵌套 cache_creation 对象
@@ -6394,6 +6513,7 @@ func (s *GatewayService) getUserGroupRateMultiplier(ctx context.Context, userID,
 // RecordUsageInput 记录使用量的输入参数
 type RecordUsageInput struct {
 	Result            *ForwardResult
+	ParsedRequest     *ParsedRequest
 	APIKey            *APIKey
 	User              *User
 	Account           *Account
@@ -6510,9 +6630,19 @@ func (s *GatewayService) RecordUsage(ctx context.Context, input *RecordUsageInpu
 		result.Usage.InputTokens = 0
 	}
 
+	// Claude Max cache billing policy (group-level):
+	// - GatewayService 路径: Forward 已改写 usage（含 cache tokens）→ apply 见到 cache tokens 跳过 → simulatedClaudeMax=true（通过第二条件）
+	// - Antigravity 路径: Forward 中 hook 改写了客户端 SSE，但 ForwardResult.Usage 是原始值 → apply 实际执行模拟 → simulatedClaudeMax=true
+	var apiKeyGroup *Group
+	if apiKey != nil {
+		apiKeyGroup = apiKey.Group
+	}
+	claudeMaxOutcome := applyClaudeMaxCacheBillingPolicyToUsage(&result.Usage, input.ParsedRequest, apiKeyGroup, result.Model, account.ID)
+	simulatedClaudeMax := claudeMaxOutcome.Simulated ||
+		(shouldApplyClaudeMaxBillingRulesForUsage(apiKeyGroup, result.Model, input.ParsedRequest) && hasCacheCreationTokens(result.Usage))
 	// Cache TTL Override: 确保计费时 token 分类与账号设置一致
 	cacheTTLOverridden := false
-	if account.IsCacheTTLOverrideEnabled() {
+	if account.IsCacheTTLOverrideEnabled() && !simulatedClaudeMax {
 		applyCacheTTLOverride(&result.Usage, account.GetCacheTTLOverrideTarget())
 		cacheTTLOverridden = (result.Usage.CacheCreation5mTokens + result.Usage.CacheCreation1hTokens) > 0
 	}

backend/internal/service/gemini_multiplatform_test.go

@@ -288,6 +288,16 @@ func (m *mockGatewayCacheForGemini) DeleteSessionAccountID(ctx context.Context,
 	return nil
 }
 
+func (m *mockGatewayCacheForGemini) GetClientAffinityAccounts(_ context.Context, _ int64, _ string, _ time.Duration) ([]int64, error) {
+	return nil, nil
+}
+func (m *mockGatewayCacheForGemini) UpdateClientAffinity(_ context.Context, _ int64, _ string, _ int64, _ time.Duration) error {
+	return nil
+}
+func (m *mockGatewayCacheForGemini) RemoveClientAffinity(_ context.Context, _ int64, _ string, _ int64) error {
+	return nil
+}
+
 // TestGeminiMessagesCompatService_SelectAccountForModelWithExclusions_GeminiPlatform 测试 Gemini 单平台选择
 func TestGeminiMessagesCompatService_SelectAccountForModelWithExclusions_GeminiPlatform(t *testing.T) {
 	ctx := context.Background()

backend/internal/service/group.go

@@ -50,6 +50,9 @@ type Group struct {
 	// MCP XML 协议注入开关（仅 antigravity 平台使用）
 	MCPXMLInject bool
 
+	// Claude usage 模拟开关：将无写缓存 usage 模拟为 claude-max 风格
+	SimulateClaudeMaxEnabled bool
+
 	// 支持的模型系列（仅 antigravity 平台使用）
 	// 可选值: claude, gemini_text, gemini_image
 	SupportedModelScopes []string

backend/internal/service/openai_gateway_service_test.go

@@ -265,6 +265,16 @@ func (c *stubGatewayCache) DeleteSessionAccountID(ctx context.Context, groupID i
 	return nil
 }
 
+func (c *stubGatewayCache) GetClientAffinityAccounts(_ context.Context, _ int64, _ string, _ time.Duration) ([]int64, error) {
+	return nil, nil
+}
+func (c *stubGatewayCache) UpdateClientAffinity(_ context.Context, _ int64, _ string, _ int64, _ time.Duration) error {
+	return nil
+}
+func (c *stubGatewayCache) RemoveClientAffinity(_ context.Context, _ int64, _ string, _ int64) error {
+	return nil
+}
+
 func TestOpenAISelectAccountWithLoadAwareness_FiltersUnschedulable(t *testing.T) {
 	now := time.Now()
 	resetAt := now.Add(10 * time.Minute)

backend/internal/service/openai_ws_state_store_test.go

@@ -193,6 +193,16 @@ func (c *openAIWSStateStoreTimeoutProbeCache) DeleteSessionAccountID(ctx context
 	return nil
 }
 
+func (c *openAIWSStateStoreTimeoutProbeCache) GetClientAffinityAccounts(_ context.Context, _ int64, _ string, _ time.Duration) ([]int64, error) {
+	return nil, nil
+}
+func (c *openAIWSStateStoreTimeoutProbeCache) UpdateClientAffinity(_ context.Context, _ int64, _ string, _ int64, _ time.Duration) error {
+	return nil
+}
+func (c *openAIWSStateStoreTimeoutProbeCache) RemoveClientAffinity(_ context.Context, _ int64, _ string, _ int64) error {
+	return nil
+}
+
 func TestOpenAIWSStateStore_RedisOpsUseShortTimeout(t *testing.T) {
 	probe := &openAIWSStateStoreTimeoutProbeCache{}
 	store := NewOpenAIWSStateStore(probe)

backend/internal/service/ops_system_log_sink_test.go

@@ -183,6 +183,15 @@ func TestOpsSystemLogSink_StartStopAndFlushSuccess(t *testing.T) {
 	if strings.TrimSpace(item.Message) == "" {
 		t.Fatalf("message should not be empty")
 	}
+	// writtenCount is incremented after BatchInsertSystemLogsFn returns,
+	// so poll briefly to avoid a race between the done signal and the atomic add.
+	deadline := time.Now().Add(time.Second)
+	for time.Now().Before(deadline) {
+		if sink.Health().WrittenCount > 0 {
+			break
+		}
+		time.Sleep(time.Millisecond)
+	}
 	health := sink.Health()
 	if health.WrittenCount == 0 {
 		t.Fatalf("written_count should be >0")

backend/internal/service/subscription_calculate_progress_test.go

@@ -34,9 +34,10 @@ func TestCalculateProgress_BasicFields(t *testing.T) {
 	assert.Equal(t, int64(100), progress.ID)
 	assert.Equal(t, "Premium", progress.GroupName)
 	assert.Equal(t, sub.ExpiresAt, progress.ExpiresAt)
-	assert.True(t, progress.ExpiresInDays == 29 || progress.ExpiresInDays == 30, "ExpiresInDays should be 29 or 30, got %d", progress.ExpiresInDays)
-	assert.Nil(t, progress.Daily, "无日限额时 Daily 应为 nil")
-	assert.Nil(t, progress.Weekly, "无周限额时 Weekly 应为 nil")
+	assert.GreaterOrEqual(t, progress.ExpiresInDays, 29)
+	assert.LessOrEqual(t, progress.ExpiresInDays, 30)
+	assert.Nil(t, progress.Daily)
+	assert.Nil(t, progress.Weekly)
 	assert.Nil(t, progress.Monthly, "无月限额时 Monthly 应为 nil")
 }
 

backend/internal/testutil/stubs.go

@@ -97,6 +97,15 @@ func (c StubGatewayCache) RefreshSessionTTL(_ context.Context, _ int64, _ string
 func (c StubGatewayCache) DeleteSessionAccountID(_ context.Context, _ int64, _ string) error {
 	return nil
 }
+func (c StubGatewayCache) GetClientAffinityAccounts(_ context.Context, _ int64, _ string, _ time.Duration) ([]int64, error) {
+	return nil, nil
+}
+func (c StubGatewayCache) UpdateClientAffinity(_ context.Context, _ int64, _ string, _ int64, _ time.Duration) error {
+	return nil
+}
+func (c StubGatewayCache) RemoveClientAffinity(_ context.Context, _ int64, _ string, _ int64) error {
+	return nil
+}
 
 // ============================================================
 // StubSessionLimitCache — service.SessionLimitCache 的空实现

backend/migrations/056_add_sonnet46_to_model_mapping.sql

@@ -0,0 +1,42 @@
+-- Add claude-sonnet-4-6 to model_mapping for all Antigravity accounts
+--
+-- Background:
+-- Antigravity now supports claude-sonnet-4-6
+--
+-- Strategy:
+-- Directly overwrite the entire model_mapping with updated mappings
+-- This ensures consistency with DefaultAntigravityModelMapping in constants.go
+
+UPDATE accounts
+SET credentials = jsonb_set(
+    credentials,
+    '{model_mapping}',
+    '{
+        "claude-opus-4-6-thinking": "claude-opus-4-6-thinking",
+        "claude-opus-4-6": "claude-opus-4-6-thinking",
+        "claude-opus-4-5-thinking": "claude-opus-4-6-thinking",
+        "claude-opus-4-5-20251101": "claude-opus-4-6-thinking",
+        "claude-sonnet-4-6": "claude-sonnet-4-6",
+        "claude-sonnet-4-5": "claude-sonnet-4-5",
+        "claude-sonnet-4-5-thinking": "claude-sonnet-4-5-thinking",
+        "claude-sonnet-4-5-20250929": "claude-sonnet-4-5",
+        "claude-haiku-4-5": "claude-sonnet-4-5",
+        "claude-haiku-4-5-20251001": "claude-sonnet-4-5",
+        "gemini-2.5-flash": "gemini-2.5-flash",
+        "gemini-2.5-flash-lite": "gemini-2.5-flash-lite",
+        "gemini-2.5-flash-thinking": "gemini-2.5-flash-thinking",
+        "gemini-2.5-pro": "gemini-2.5-pro",
+        "gemini-3-flash": "gemini-3-flash",
+        "gemini-3-pro-high": "gemini-3-pro-high",
+        "gemini-3-pro-low": "gemini-3-pro-low",
+        "gemini-3-pro-image": "gemini-3-pro-image",
+        "gemini-3-flash-preview": "gemini-3-flash",
+        "gemini-3-pro-preview": "gemini-3-pro-high",
+        "gemini-3-pro-image-preview": "gemini-3-pro-image",
+        "gpt-oss-120b-medium": "gpt-oss-120b-medium",
+        "tab_flash_lite_preview": "tab_flash_lite_preview"
+    }'::jsonb
+)
+WHERE platform = 'antigravity'
+  AND deleted_at IS NULL
+  AND credentials->'model_mapping' IS NOT NULL;

backend/migrations/057_add_gemini31_pro_to_model_mapping.sql

@@ -0,0 +1,45 @@
+-- Add gemini-3.1-pro-high, gemini-3.1-pro-low, gemini-3.1-pro-preview to model_mapping
+--
+-- Background:
+-- Antigravity now supports gemini-3.1-pro-high and gemini-3.1-pro-low
+--
+-- Strategy:
+-- Directly overwrite the entire model_mapping with updated mappings
+-- This ensures consistency with DefaultAntigravityModelMapping in constants.go
+
+UPDATE accounts
+SET credentials = jsonb_set(
+    credentials,
+    '{model_mapping}',
+    '{
+        "claude-opus-4-6-thinking": "claude-opus-4-6-thinking",
+        "claude-opus-4-6": "claude-opus-4-6-thinking",
+        "claude-opus-4-5-thinking": "claude-opus-4-6-thinking",
+        "claude-opus-4-5-20251101": "claude-opus-4-6-thinking",
+        "claude-sonnet-4-6": "claude-sonnet-4-6",
+        "claude-sonnet-4-5": "claude-sonnet-4-5",
+        "claude-sonnet-4-5-thinking": "claude-sonnet-4-5-thinking",
+        "claude-sonnet-4-5-20250929": "claude-sonnet-4-5",
+        "claude-haiku-4-5": "claude-sonnet-4-5",
+        "claude-haiku-4-5-20251001": "claude-sonnet-4-5",
+        "gemini-2.5-flash": "gemini-2.5-flash",
+        "gemini-2.5-flash-lite": "gemini-2.5-flash-lite",
+        "gemini-2.5-flash-thinking": "gemini-2.5-flash-thinking",
+        "gemini-2.5-pro": "gemini-2.5-pro",
+        "gemini-3-flash": "gemini-3-flash",
+        "gemini-3-pro-high": "gemini-3-pro-high",
+        "gemini-3-pro-low": "gemini-3-pro-low",
+        "gemini-3-pro-image": "gemini-3-pro-image",
+        "gemini-3-flash-preview": "gemini-3-flash",
+        "gemini-3-pro-preview": "gemini-3-pro-high",
+        "gemini-3-pro-image-preview": "gemini-3-pro-image",
+        "gemini-3.1-pro-high": "gemini-3.1-pro-high",
+        "gemini-3.1-pro-low": "gemini-3.1-pro-low",
+        "gemini-3.1-pro-preview": "gemini-3.1-pro-high",
+        "gpt-oss-120b-medium": "gpt-oss-120b-medium",
+        "tab_flash_lite_preview": "tab_flash_lite_preview"
+    }'::jsonb
+)
+WHERE platform = 'antigravity'
+  AND deleted_at IS NULL
+  AND credentials->'model_mapping' IS NOT NULL;

backend/migrations/060_add_group_simulate_claude_max.sql

@@ -0,0 +1,3 @@
+ALTER TABLE groups
+    ADD COLUMN IF NOT EXISTS simulate_claude_max_enabled BOOLEAN NOT NULL DEFAULT FALSE;
+

deploy/docker-compose.yml

@@ -47,13 +47,15 @@ services:
 
       # =======================================================================
       # Database Configuration (PostgreSQL)
+      # Default: uses local postgres container
+      # External DB: set DATABASE_HOST and DATABASE_SSLMODE in .env
       # =======================================================================
-      - DATABASE_HOST=postgres
-      - DATABASE_PORT=5432
+      - DATABASE_HOST=${DATABASE_HOST:-postgres}
+      - DATABASE_PORT=${DATABASE_PORT:-5432}
       - DATABASE_USER=${POSTGRES_USER:-sub2api}
       - DATABASE_PASSWORD=${POSTGRES_PASSWORD:?POSTGRES_PASSWORD is required}
       - DATABASE_DBNAME=${POSTGRES_DB:-sub2api}
-      - DATABASE_SSLMODE=disable
+      - DATABASE_SSLMODE=${DATABASE_SSLMODE:-disable}
       - DATABASE_MAX_OPEN_CONNS=${DATABASE_MAX_OPEN_CONNS:-50}
       - DATABASE_MAX_IDLE_CONNS=${DATABASE_MAX_IDLE_CONNS:-10}
       - DATABASE_CONN_MAX_LIFETIME_MINUTES=${DATABASE_CONN_MAX_LIFETIME_MINUTES:-30}
@@ -139,8 +141,6 @@ services:
       # Examples: http://host:port, socks5://host:port
       - UPDATE_PROXY_URL=${UPDATE_PROXY_URL:-}
     depends_on:
-      postgres:
-        condition: service_healthy
       redis:
         condition: service_healthy
     networks:

frontend/src/components/account/AccountStatusIndicator.vue

@@ -81,15 +81,15 @@
       v-if="activeModelRateLimits.length > 0"
       :class="[
         activeModelRateLimits.length <= 4
-          ? 'flex flex-col gap-1'
+          ? 'flex flex-col gap-0.5'
           : activeModelRateLimits.length <= 8
             ? 'columns-2 gap-x-2'
             : 'columns-3 gap-x-2'
       ]"
     >
-      <div v-for="item in activeModelRateLimits" :key="item.model" class="group relative mb-1 break-inside-avoid">
+      <div v-for="item in activeModelRateLimits" :key="item.model" class="group relative mb-0.5 break-inside-avoid">
         <span
-          class="inline-flex items-center gap-1 rounded bg-purple-100 px-1.5 py-0.5 text-xs font-medium text-purple-700 dark:bg-purple-900/30 dark:text-purple-400"
+          class="inline-flex items-center gap-0.5 rounded bg-purple-100 px-1.5 py-px text-xs font-medium text-purple-700 dark:bg-purple-900/30 dark:text-purple-400"
         >
           <Icon name="exclamationTriangle" size="xs" :stroke-width="2" />
           {{ formatScopeName(item.model) }}

frontend/src/components/account/AccountUsageCell.vue

@@ -278,11 +278,12 @@
 </template>
 
 <script setup lang="ts">
-import { ref, computed, onMounted } from 'vue'
+import { ref, computed, onMounted, onBeforeUnmount } from 'vue'
 import { useI18n } from 'vue-i18n'
 import { adminAPI } from '@/api/admin'
 import type { Account, AccountUsageInfo, GeminiCredentials, WindowStats } from '@/types'
 import { resolveCodexUsageWindow } from '@/utils/codexUsage'
+import { enqueueUsageRequest } from '@/utils/usageLoadQueue'
 import UsageProgressBar from './UsageProgressBar.vue'
 import AccountQuotaInfo from './AccountQuotaInfo.vue'
 
@@ -292,6 +293,9 @@ const props = defineProps<{
 
 const { t } = useI18n()
 
+const unmounted = ref(false)
+onBeforeUnmount(() => { unmounted.value = true })
+
 const loading = ref(false)
 const error = ref<string | null>(null)
 const usageInfo = ref<AccountUsageInfo | null>(null)
@@ -701,12 +705,30 @@ const loadUsage = async () => {
   error.value = null
 
   try {
-    usageInfo.value = await adminAPI.accounts.getUsage(props.account.id)
+    const fetchFn = () => adminAPI.accounts.getUsage(props.account.id)
+    let result: AccountUsageInfo
+    // Only throttle Anthropic OAuth/setup-token accounts to avoid upstream 429
+    if (
+      props.account.platform === 'anthropic' &&
+      (props.account.type === 'oauth' || props.account.type === 'setup-token')
+    ) {
+      result = await enqueueUsageRequest(
+        props.account.platform,
+        'claude_code',
+        props.account.proxy_id,
+        fetchFn
+      )
+    } else {
+      result = await fetchFn()
+    }
+    if (!unmounted.value) usageInfo.value = result
   } catch (e: any) {
-    error.value = t('common.error')
-    console.error('Failed to load usage:', e)
+    if (!unmounted.value) {
+      error.value = t('common.error')
+      console.error('Failed to load usage:', e)
+    }
   } finally {
-    loading.value = false
+    if (!unmounted.value) loading.value = false
   }
 }
 

frontend/src/components/account/BulkEditAccountModal.vue

@@ -5,7 +5,7 @@
     width="wide"
     @close="handleClose"
   >
-    <form id="bulk-edit-account-form" class="space-y-5" @submit.prevent="handleSubmit">
+    <form id="bulk-edit-account-form" class="space-y-5" @submit.prevent="() => handleSubmit()">
       <!-- Info -->
       <div class="rounded-lg bg-blue-50 p-4 dark:bg-blue-900/20">
         <p class="text-sm text-blue-700 dark:text-blue-400">

frontend/src/components/account/EditAccountModal.vue

@@ -1165,6 +1165,38 @@
         </div>
       </div>
 
+      <!-- Client Affinity (Anthropic/Antigravity platform) -->
+      <div
+        v-if="(account?.platform === 'anthropic' && (account?.type === 'oauth' || account?.type === 'setup-token')) || account?.platform === 'antigravity'"
+        class="border-t border-gray-200 pt-4 dark:border-dark-600"
+      >
+        <div class="rounded-lg border border-gray-200 p-4 dark:border-dark-600">
+          <div class="flex items-center justify-between">
+            <div>
+              <label class="input-label mb-0">{{ t('admin.accounts.quotaControl.clientAffinity.label') }}</label>
+              <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                {{ t('admin.accounts.quotaControl.clientAffinity.hint') }}
+              </p>
+            </div>
+            <button
+              type="button"
+              @click="clientAffinityEnabled = !clientAffinityEnabled"
+              :class="[
+                'relative inline-flex h-6 w-11 flex-shrink-0 cursor-pointer rounded-full border-2 border-transparent transition-colors duration-200 ease-in-out focus:outline-none focus:ring-2 focus:ring-primary-500 focus:ring-offset-2',
+                clientAffinityEnabled ? 'bg-primary-600' : 'bg-gray-200 dark:bg-dark-600'
+              ]"
+            >
+              <span
+                :class="[
+                  'pointer-events-none inline-block h-5 w-5 transform rounded-full bg-white shadow ring-0 transition duration-200 ease-in-out',
+                  clientAffinityEnabled ? 'translate-x-5' : 'translate-x-0'
+                ]"
+              />
+            </button>
+          </div>
+        </div>
+      </div>
+
       <div class="border-t border-gray-200 pt-4 dark:border-dark-600">
         <div>
           <label class="input-label">{{ t('common.status') }}</label>
@@ -1391,6 +1423,7 @@ const tlsFingerprintEnabled = ref(false)
 const sessionIdMaskingEnabled = ref(false)
 const cacheTTLOverrideEnabled = ref(false)
 const cacheTTLOverrideTarget = ref<string>('5m')
+const clientAffinityEnabled = ref(false)
 
 // OpenAI 自动透传开关（OAuth/API Key）
 const openaiPassthroughEnabled = ref(false)
@@ -1872,6 +1905,12 @@ function loadQuotaControlSettings(account: Account) {
   sessionIdMaskingEnabled.value = false
   cacheTTLOverrideEnabled.value = false
   cacheTTLOverrideTarget.value = '5m'
+  clientAffinityEnabled.value = false
+
+  // Client affinity applies to both Anthropic and Antigravity platforms
+  if (account.client_affinity_enabled === true) {
+    clientAffinityEnabled.value = true
+  }
 
   // Only applies to Anthropic OAuth/SetupToken accounts
   if (account.platform !== 'anthropic' || (account.type !== 'oauth' && account.type !== 'setup-token')) {
@@ -2196,6 +2235,12 @@ const handleSubmit = async () => {
       } else {
         delete newExtra.mixed_scheduling
       }
+      // Client affinity setting
+      if (clientAffinityEnabled.value) {
+        newExtra.client_affinity_enabled = true
+      } else {
+        delete newExtra.client_affinity_enabled
+      }
       updatePayload.extra = newExtra
     }
 
@@ -2271,6 +2316,13 @@ const handleSubmit = async () => {
         delete newExtra.cache_ttl_override_target
       }
 
+      // Client affinity setting
+      if (clientAffinityEnabled.value) {
+        newExtra.client_affinity_enabled = true
+      } else {
+        delete newExtra.client_affinity_enabled
+      }
+
       updatePayload.extra = newExtra
     }
 

frontend/src/components/account/__tests__/AccountUsageCell.spec.ts

@@ -14,6 +14,10 @@ vi.mock('@/api/admin', () => ({
   }
 }))
 
+vi.mock('@/utils/usageLoadQueue', () => ({
+  enqueueUsageRequest: (_p: string, _t: string, _id: unknown, fn: () => Promise<unknown>) => fn()
+}))
+
 vi.mock('vue-i18n', async () => {
   const actual = await vi.importActual<typeof import('vue-i18n')>('vue-i18n')
   return {

frontend/src/components/common/WechatServiceButton.vue

@@ -0,0 +1,104 @@
+<template>
+  <!-- 悬浮按钮 - 使用主题色 -->
+  <button
+    @click="showModal = true"
+    class="fixed bottom-6 right-6 z-50 flex items-center gap-2 rounded-full bg-gradient-to-r from-primary-500 to-primary-600 px-4 py-3 text-white shadow-lg shadow-primary-500/25 transition-all hover:from-primary-600 hover:to-primary-700 hover:shadow-xl hover:shadow-primary-500/30"
+  >
+    <svg class="h-5 w-5" viewBox="0 0 24 24" fill="currentColor">
+      <path d="M8.691 2.188C3.891 2.188 0 5.476 0 9.53c0 2.212 1.17 4.203 3.002 5.55a.59.59 0 01.213.665l-.39 1.48c-.019.07-.048.141-.048.213 0 .163.13.295.29.295a.328.328 0 00.186-.059l2.114-1.225a.87.87 0 01.415-.106.807.807 0 01.213.026 10.07 10.07 0 002.696.37c.262 0 .52-.011.776-.028a5.91 5.91 0 01-.193-1.479c0-3.644 3.374-6.6 7.536-6.6.262 0 .52.011.776.028-.628-3.513-4.27-6.472-8.885-6.472zM5.785 5.97a1.1 1.1 0 110 2.2 1.1 1.1 0 010-2.2zm5.813 0a1.1 1.1 0 110 2.2 1.1 1.1 0 010-2.2zm5.192 2.642c-3.703 0-6.71 2.567-6.71 5.73 0 3.163 3.007 5.73 6.71 5.73a7.9 7.9 0 002.126-.288.644.644 0 01.17-.022.69.69 0 01.329.085l1.672.97a.262.262 0 00.147.046c.128 0 .23-.104.23-.233a.403.403 0 00-.038-.168l-.309-1.17a.468.468 0 01.168-.527c1.449-1.065 2.374-2.643 2.374-4.423 0-3.163-3.007-5.73-6.71-5.73h-.159zm-2.434 3.34a.88.88 0 110 1.76.88.88 0 010-1.76zm4.868 0a.88.88 0 110 1.76.88.88 0 010-1.76z"/>
+    </svg>
+    <span class="text-sm font-medium">客服</span>
+  </button>
+
+  <!-- 弹窗 -->
+  <Teleport to="body">
+    <Transition name="fade">
+      <div
+        v-if="showModal"
+        class="fixed inset-0 z-[100] flex items-center justify-center bg-black/50 p-4 backdrop-blur-sm"
+        @click.self="showModal = false"
+      >
+        <Transition name="scale">
+          <div
+            v-if="showModal"
+            class="relative w-full max-w-sm rounded-2xl bg-white p-6 shadow-2xl dark:bg-dark-700"
+          >
+            <!-- 关闭按钮 -->
+            <button
+              @click="showModal = false"
+              class="absolute right-4 top-4 text-gray-400 transition-colors hover:text-gray-600 dark:text-dark-400 dark:hover:text-dark-200"
+            >
+              <svg class="h-5 w-5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
+                <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M6 18L18 6M6 6l12 12" />
+              </svg>
+            </button>
+
+            <!-- 标题 -->
+            <div class="mb-4 flex items-center gap-3">
+              <div class="flex h-10 w-10 items-center justify-center rounded-full bg-gradient-to-br from-primary-500 to-primary-600">
+                <svg class="h-6 w-6 text-white" viewBox="0 0 24 24" fill="currentColor">
+                  <path d="M8.691 2.188C3.891 2.188 0 5.476 0 9.53c0 2.212 1.17 4.203 3.002 5.55a.59.59 0 01.213.665l-.39 1.48c-.019.07-.048.141-.048.213 0 .163.13.295.29.295a.328.328 0 00.186-.059l2.114-1.225a.87.87 0 01.415-.106.807.807 0 01.213.026 10.07 10.07 0 002.696.37c.262 0 .52-.011.776-.028a5.91 5.91 0 01-.193-1.479c0-3.644 3.374-6.6 7.536-6.6.262 0 .52.011.776.028-.628-3.513-4.27-6.472-8.885-6.472zM5.785 5.97a1.1 1.1 0 110 2.2 1.1 1.1 0 010-2.2zm5.813 0a1.1 1.1 0 110 2.2 1.1 1.1 0 010-2.2zm5.192 2.642c-3.703 0-6.71 2.567-6.71 5.73 0 3.163 3.007 5.73 6.71 5.73a7.9 7.9 0 002.126-.288.644.644 0 01.17-.022.69.69 0 01.329.085l1.672.97a.262.262 0 00.147.046c.128 0 .23-.104.23-.233a.403.403 0 00-.038-.168l-.309-1.17a.468.468 0 01.168-.527c1.449-1.065 2.374-2.643 2.374-4.423 0-3.163-3.007-5.73-6.71-5.73h-.159zm-2.434 3.34a.88.88 0 110 1.76.88.88 0 010-1.76zm4.868 0a.88.88 0 110 1.76.88.88 0 010-1.76z"/>
+                </svg>
+              </div>
+              <div>
+                <h3 class="text-lg font-semibold text-gray-900 dark:text-white">联系客服</h3>
+                <p class="text-sm text-gray-500 dark:text-dark-400">扫码添加好友</p>
+              </div>
+            </div>
+
+            <!-- 二维码卡片 -->
+            <div class="mb-4 overflow-hidden rounded-xl border border-primary-100 bg-gradient-to-br from-primary-50 to-white p-3 dark:border-primary-800/30 dark:from-primary-900/10 dark:to-dark-800">
+              <img
+                src="/wechat-qr.jpg"
+                alt="微信二维码"
+                class="w-full rounded-lg"
+              />
+            </div>
+
+            <!-- 提示文字 -->
+            <div class="text-center">
+              <p class="mb-2 text-sm font-medium text-primary-600 dark:text-primary-400">
+                微信扫码添加客服
+              </p>
+              <p class="flex items-center justify-center gap-1 text-xs text-gray-500 dark:text-dark-400">
+                <svg class="h-4 w-4" fill="none" stroke="currentColor" viewBox="0 0 24 24">
+                  <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M12 8v4l3 3m6-3a9 9 0 11-18 0 9 9 0 0118 0z" />
+                </svg>
+                工作时间：周一至周五 9:00-18:00
+              </p>
+            </div>
+          </div>
+        </Transition>
+      </div>
+    </Transition>
+  </Teleport>
+</template>
+
+<script setup lang="ts">
+import { ref } from 'vue'
+
+const showModal = ref(false)
+</script>
+
+<style scoped>
+.fade-enter-active,
+.fade-leave-active {
+  transition: opacity 0.2s ease;
+}
+
+.fade-enter-from,
+.fade-leave-to {
+  opacity: 0;
+}
+
+.scale-enter-active,
+.scale-leave-active {
+  transition: all 0.2s ease;
+}
+
+.scale-enter-from,
+.scale-leave-to {
+  opacity: 0;
+  transform: scale(0.95);
+}
+</style>

frontend/src/components/layout/AppHeader.vue

@@ -121,23 +121,6 @@
                   <Icon name="key" size="sm" />
                   {{ t('nav.apiKeys') }}
                 </router-link>
-
-                <a
-                  href="https://github.com/Wei-Shaw/sub2api"
-                  target="_blank"
-                  rel="noopener noreferrer"
-                  @click="closeDropdown"
-                  class="dropdown-item"
-                >
-                  <svg class="h-4 w-4" fill="currentColor" viewBox="0 0 24 24">
-                    <path
-                      fill-rule="evenodd"
-                      clip-rule="evenodd"
-                      d="M12 2C6.477 2 2 6.477 2 12c0 4.42 2.865 8.17 6.839 9.49.5.092.682-.217.682-.482 0-.237-.008-.866-.013-1.7-2.782.604-3.369-1.34-3.369-1.34-.454-1.156-1.11-1.464-1.11-1.464-.908-.62.069-.608.069-.608 1.003.07 1.531 1.03 1.531 1.03.892 1.529 2.341 1.087 2.91.831.092-.646.35-1.086.636-1.336-2.22-.253-4.555-1.11-4.555-4.943 0-1.091.39-1.984 1.029-2.683-.103-.253-.446-1.27.098-2.647 0 0 .84-.269 2.75 1.025A9.578 9.578 0 0112 6.836c.85.004 1.705.114 2.504.336 1.909-1.294 2.747-1.025 2.747-1.025.546 1.377.203 2.394.1 2.647.64.699 1.028 1.592 1.028 2.683 0 3.842-2.339 4.687-4.566 4.935.359.309.678.919.678 1.852 0 1.336-.012 2.415-.012 2.743 0 .267.18.578.688.48C19.138 20.167 22 16.418 22 12c0-5.523-4.477-10-10-10z"
-                    />
-                  </svg>
-                  {{ t('nav.github') }}
-                </a>
               </div>
 
               <!-- Contact Support (only show if configured) -->

frontend/src/i18n/locales/en.ts

@@ -1482,6 +1482,14 @@ export default {
         enabled: 'Enabled',
         disabled: 'Disabled'
       },
+      claudeMaxSimulation: {
+        title: 'Claude Max Usage Simulation',
+        tooltip:
+          'When enabled, for Claude models without upstream cache-write usage, the system deterministically maps tokens to a small input plus 1h cache creation while keeping total tokens unchanged.',
+        enabled: 'Enabled (simulate 1h cache)',
+        disabled: 'Disabled',
+        hint: 'Only token categories in usage billing logs are adjusted. No per-request mapping state is persisted.'
+      },
       supportedScopes: {
         title: 'Supported Model Families',
         tooltip: 'Select the model families this group supports. Unchecked families will not be routed to this group.',
@@ -1997,6 +2005,10 @@ export default {
           hint: 'Force all cache creation tokens to be billed as the selected TTL tier (5m or 1h)',
           target: 'Target TTL',
           targetHint: 'Select the TTL tier for billing'
+        },
+        clientAffinity: {
+          label: 'Client Affinity Scheduling',
+          hint: 'When enabled, new sessions prefer accounts previously used by this client to reduce account switching'
         }
       },
       expired: 'Expired',

frontend/src/i18n/locales/zh.ts

@@ -2139,6 +2139,10 @@ export default {
           hint: '将所有缓存创建 token 强制按指定的 TTL 类型（5分钟或1小时）计费',
           target: '目标 TTL',
           targetHint: '选择计费使用的 TTL 类型'
+        },
+        clientAffinity: {
+          label: '客户端亲和调度',
+          hint: '启用后，新会话会优先调度到该客户端之前使用过的账号，避免频繁切换账号'
         }
       },
       expired: '已过期',

frontend/src/types/index.ts

@@ -395,6 +395,8 @@ export interface AdminGroup extends Group {
 
   // MCP XML 协议注入（仅 antigravity 平台使用）
   mcp_xml_inject: boolean
+  // Claude usage 模拟开关（仅 anthropic 平台使用）
+  simulate_claude_max_enabled: boolean
 
   // 支持的模型系列（仅 antigravity 平台使用）
   supported_model_scopes?: string[]
@@ -483,6 +485,7 @@ export interface CreateGroupRequest {
   fallback_group_id?: number | null
   fallback_group_id_on_invalid_request?: number | null
   mcp_xml_inject?: boolean
+  simulate_claude_max_enabled?: boolean
   supported_model_scopes?: string[]
   // 从指定分组复制账号
   copy_accounts_from_group_ids?: number[]
@@ -511,6 +514,7 @@ export interface UpdateGroupRequest {
   fallback_group_id?: number | null
   fallback_group_id_on_invalid_request?: number | null
   mcp_xml_inject?: boolean
+  simulate_claude_max_enabled?: boolean
   supported_model_scopes?: string[]
   copy_accounts_from_group_ids?: number[]
 }
@@ -706,6 +710,10 @@ export interface Account {
   cache_ttl_override_enabled?: boolean | null
   cache_ttl_override_target?: string | null
 
+  // 客户端亲和调度（仅 Anthropic/Antigravity 平台有效）
+  // 启用后新会话会优先调度到客户端之前使用过的账号
+  client_affinity_enabled?: boolean | null
+
   // API Key 账号配额限制
   quota_limit?: number | null
   quota_used?: number | null

frontend/src/utils/__tests__/usageLoadQueue.spec.ts

@@ -0,0 +1,87 @@
+import { describe, expect, it, vi } from 'vitest'
+import { enqueueUsageRequest } from '../usageLoadQueue'
+
+function delay(ms: number) {
+  return new Promise((r) => setTimeout(r, ms))
+}
+
+describe('usageLoadQueue', () => {
+  it('同组请求串行执行，间隔 >= 1s', async () => {
+    const timestamps: number[] = []
+    const makeFn = () => async () => {
+      timestamps.push(Date.now())
+      return 'ok'
+    }
+
+    const p1 = enqueueUsageRequest('anthropic', 'oauth', 1, makeFn())
+    const p2 = enqueueUsageRequest('anthropic', 'oauth', 1, makeFn())
+    const p3 = enqueueUsageRequest('anthropic', 'oauth', 1, makeFn())
+
+    await Promise.all([p1, p2, p3])
+
+    expect(timestamps).toHaveLength(3)
+    // 随机 1-1.5s 间隔，至少 950ms（留一点误差）
+    expect(timestamps[1] - timestamps[0]).toBeGreaterThanOrEqual(950)
+    expect(timestamps[1] - timestamps[0]).toBeLessThan(1600)
+    expect(timestamps[2] - timestamps[1]).toBeGreaterThanOrEqual(950)
+    expect(timestamps[2] - timestamps[1]).toBeLessThan(1600)
+  })
+
+  it('不同组请求并行执行', async () => {
+    const timestamps: Record<string, number> = {}
+    const makeTracked = (key: string) => async () => {
+      timestamps[key] = Date.now()
+      return key
+    }
+
+    const p1 = enqueueUsageRequest('anthropic', 'oauth', 1, makeTracked('group1'))
+    const p2 = enqueueUsageRequest('anthropic', 'oauth', 2, makeTracked('group2'))
+    const p3 = enqueueUsageRequest('gemini', 'oauth', 1, makeTracked('group3'))
+
+    await Promise.all([p1, p2, p3])
+
+    // 不同组应几乎同时启动（差距 < 50ms）
+    const values = Object.values(timestamps)
+    const spread = Math.max(...values) - Math.min(...values)
+    expect(spread).toBeLessThan(50)
+  })
+
+  it('请求失败时 reject，后续任务继续执行', async () => {
+    const results: string[] = []
+
+    const p1 = enqueueUsageRequest('anthropic', 'oauth', 99, async () => {
+      throw new Error('fail')
+    })
+    const p2 = enqueueUsageRequest('anthropic', 'oauth', 99, async () => {
+      results.push('second')
+      return 'ok'
+    })
+
+    await expect(p1).rejects.toThrow('fail')
+    await p2
+    expect(results).toEqual(['second'])
+  })
+
+  it('返回值正确透传', async () => {
+    const result = await enqueueUsageRequest('test', 'oauth', null, async () => {
+      return { usage: 42 }
+    })
+    expect(result).toEqual({ usage: 42 })
+  })
+
+  it('proxy_id 为 null 的账号归为同一组', async () => {
+    const order: number[] = []
+    const makeFn = (n: number) => async () => {
+      order.push(n)
+      return n
+    }
+
+    const p1 = enqueueUsageRequest('anthropic', 'oauth', null, makeFn(1))
+    const p2 = enqueueUsageRequest('anthropic', 'oauth', null, makeFn(2))
+
+    await Promise.all([p1, p2])
+
+    // 同组串行，按入队顺序执行
+    expect(order).toEqual([1, 2])
+  })
+})

frontend/src/utils/usageLoadQueue.ts

@@ -0,0 +1,72 @@
+/**
+ * Usage request queue that throttles API calls by group.
+ *
+ * Accounts sharing the same upstream (platform + type + proxy) are placed
+ * into a single serial queue with a configurable delay between requests,
+ * preventing upstream 429 rate-limit errors.
+ *
+ * Different groups run in parallel since they hit different upstreams.
+ */
+
+const GROUP_DELAY_MIN_MS = 1000
+const GROUP_DELAY_MAX_MS = 1500
+
+type Task<T> = {
+  fn: () => Promise<T>
+  resolve: (value: T) => void
+  reject: (reason: unknown) => void
+}
+
+const queues = new Map<string, Task<unknown>[]>()
+const running = new Set<string>()
+
+function buildGroupKey(platform: string, type: string, proxyId: number | null): string {
+  return `${platform}:${type}:${proxyId ?? 'direct'}`
+}
+
+async function drain(groupKey: string) {
+  if (running.has(groupKey)) return
+  running.add(groupKey)
+
+  const queue = queues.get(groupKey)
+  while (queue && queue.length > 0) {
+    const task = queue.shift()!
+    try {
+      const result = await task.fn()
+      task.resolve(result)
+    } catch (err) {
+      task.reject(err)
+    }
+    // Wait a random 1–1.5s before next request in the same group
+    if (queue.length > 0) {
+      const jitter = GROUP_DELAY_MIN_MS + Math.random() * (GROUP_DELAY_MAX_MS - GROUP_DELAY_MIN_MS)
+      await new Promise((r) => setTimeout(r, jitter))
+    }
+  }
+
+  running.delete(groupKey)
+  queues.delete(groupKey)
+}
+
+/**
+ * Enqueue a usage fetch call. Returns a promise that resolves when the
+ * request completes (after waiting its turn in the group queue).
+ */
+export function enqueueUsageRequest<T>(
+  platform: string,
+  type: string,
+  proxyId: number | null,
+  fn: () => Promise<T>
+): Promise<T> {
+  const key = buildGroupKey(platform, type, proxyId)
+
+  return new Promise<T>((resolve, reject) => {
+    let queue = queues.get(key)
+    if (!queue) {
+      queue = []
+      queues.set(key, queue)
+    }
+    queue.push({ fn, resolve, reject } as Task<unknown>)
+    drain(key)
+  })
+}

frontend/src/views/HomeView.vue

@@ -122,8 +122,11 @@
             >
               {{ siteName }}
             </h1>
-            <p class="mb-8 text-lg text-gray-600 dark:text-dark-300 md:text-xl">
-              {{ siteSubtitle }}
+            <p class="mb-3 text-xl font-semibold text-primary-600 dark:text-primary-400 md:text-2xl">
+              {{ t('home.heroSubtitle') }}
+            </p>
+            <p class="mb-8 text-base text-gray-600 dark:text-dark-300 md:text-lg">
+              {{ t('home.heroDescription') }}
             </p>
 
             <!-- CTA Button -->
@@ -177,7 +180,7 @@
         </div>
 
         <!-- Feature Tags - Centered -->
-        <div class="mb-12 flex flex-wrap items-center justify-center gap-4 md:gap-6">
+        <div class="mb-16 flex flex-wrap items-center justify-center gap-4 md:gap-6">
           <div
             class="inline-flex items-center gap-2.5 rounded-full border border-gray-200/50 bg-white/80 px-5 py-2.5 shadow-sm backdrop-blur-sm dark:border-dark-700/50 dark:bg-dark-800/80"
           >
@@ -204,6 +207,63 @@
           </div>
         </div>
 
+        <!-- Pain Points Section -->
+        <div class="mb-16">
+          <h2 class="mb-8 text-center text-2xl font-bold text-gray-900 dark:text-white md:text-3xl">
+            {{ t('home.painPoints.title') }}
+          </h2>
+          <div class="grid gap-4 sm:grid-cols-2 lg:grid-cols-4">
+            <!-- Pain Point 1: Expensive -->
+            <div class="rounded-xl border border-red-200/50 bg-red-50/50 p-5 dark:border-red-900/30 dark:bg-red-950/20">
+              <div class="mb-3 flex h-10 w-10 items-center justify-center rounded-lg bg-red-100 dark:bg-red-900/30">
+                <svg class="h-5 w-5 text-red-500" fill="none" viewBox="0 0 24 24" stroke="currentColor" stroke-width="2">
+                  <path stroke-linecap="round" stroke-linejoin="round" d="M12 8c-1.657 0-3 .895-3 2s1.343 2 3 2 3 .895 3 2-1.343 2-3 2m0-8c1.11 0 2.08.402 2.599 1M12 8V7m0 1v8m0 0v1m0-1c-1.11 0-2.08-.402-2.599-1M21 12a9 9 0 11-18 0 9 9 0 0118 0z" />
+                </svg>
+              </div>
+              <h3 class="mb-1.5 font-semibold text-gray-900 dark:text-white">{{ t('home.painPoints.items.expensive.title') }}</h3>
+              <p class="text-sm text-gray-600 dark:text-dark-400">{{ t('home.painPoints.items.expensive.desc') }}</p>
+            </div>
+            <!-- Pain Point 2: Complex -->
+            <div class="rounded-xl border border-orange-200/50 bg-orange-50/50 p-5 dark:border-orange-900/30 dark:bg-orange-950/20">
+              <div class="mb-3 flex h-10 w-10 items-center justify-center rounded-lg bg-orange-100 dark:bg-orange-900/30">
+                <svg class="h-5 w-5 text-orange-500" fill="none" viewBox="0 0 24 24" stroke="currentColor" stroke-width="2">
+                  <path stroke-linecap="round" stroke-linejoin="round" d="M19 11H5m14 0a2 2 0 012 2v6a2 2 0 01-2 2H5a2 2 0 01-2-2v-6a2 2 0 012-2m14 0V9a2 2 0 00-2-2M5 11V9a2 2 0 012-2m0 0V5a2 2 0 012-2h6a2 2 0 012 2v2M7 7h10" />
+                </svg>
+              </div>
+              <h3 class="mb-1.5 font-semibold text-gray-900 dark:text-white">{{ t('home.painPoints.items.complex.title') }}</h3>
+              <p class="text-sm text-gray-600 dark:text-dark-400">{{ t('home.painPoints.items.complex.desc') }}</p>
+            </div>
+            <!-- Pain Point 3: Unstable -->
+            <div class="rounded-xl border border-yellow-200/50 bg-yellow-50/50 p-5 dark:border-yellow-900/30 dark:bg-yellow-950/20">
+              <div class="mb-3 flex h-10 w-10 items-center justify-center rounded-lg bg-yellow-100 dark:bg-yellow-900/30">
+                <svg class="h-5 w-5 text-yellow-600" fill="none" viewBox="0 0 24 24" stroke="currentColor" stroke-width="2">
+                  <path stroke-linecap="round" stroke-linejoin="round" d="M12 9v2m0 4h.01m-6.938 4h13.856c1.54 0 2.502-1.667 1.732-3L13.732 4c-.77-1.333-2.694-1.333-3.464 0L3.34 16c-.77 1.333.192 3 1.732 3z" />
+                </svg>
+              </div>
+              <h3 class="mb-1.5 font-semibold text-gray-900 dark:text-white">{{ t('home.painPoints.items.unstable.title') }}</h3>
+              <p class="text-sm text-gray-600 dark:text-dark-400">{{ t('home.painPoints.items.unstable.desc') }}</p>
+            </div>
+            <!-- Pain Point 4: No Control -->
+            <div class="rounded-xl border border-gray-200/50 bg-gray-50/50 p-5 dark:border-dark-700/50 dark:bg-dark-800/50">
+              <div class="mb-3 flex h-10 w-10 items-center justify-center rounded-lg bg-gray-100 dark:bg-dark-700">
+                <svg class="h-5 w-5 text-gray-500" fill="none" viewBox="0 0 24 24" stroke="currentColor" stroke-width="2">
+                  <path stroke-linecap="round" stroke-linejoin="round" d="M18.364 18.364A9 9 0 005.636 5.636m12.728 12.728A9 9 0 015.636 5.636m12.728 12.728L5.636 5.636" />
+                </svg>
+              </div>
+              <h3 class="mb-1.5 font-semibold text-gray-900 dark:text-white">{{ t('home.painPoints.items.noControl.title') }}</h3>
+              <p class="text-sm text-gray-600 dark:text-dark-400">{{ t('home.painPoints.items.noControl.desc') }}</p>
+            </div>
+          </div>
+        </div>
+
+        <!-- Solutions Section Title -->
+        <div class="mb-8 text-center">
+          <h2 class="mb-2 text-2xl font-bold text-gray-900 dark:text-white md:text-3xl">
+            {{ t('home.solutions.title') }}
+          </h2>
+          <p class="text-gray-600 dark:text-dark-400">{{ t('home.solutions.subtitle') }}</p>
+        </div>
+
         <!-- Features Grid -->
         <div class="mb-12 grid gap-6 md:grid-cols-3">
           <!-- Feature 1: Unified Gateway -->
@@ -369,6 +429,77 @@
             >
           </div>
         </div>
+
+        <!-- Comparison Table -->
+        <div class="mb-16">
+          <h2 class="mb-8 text-center text-2xl font-bold text-gray-900 dark:text-white md:text-3xl">
+            {{ t('home.comparison.title') }}
+          </h2>
+          <div class="overflow-x-auto">
+            <table class="w-full rounded-xl border border-gray-200/50 bg-white/60 backdrop-blur-sm dark:border-dark-700/50 dark:bg-dark-800/60">
+              <thead>
+                <tr class="border-b border-gray-200/50 dark:border-dark-700/50">
+                  <th class="px-6 py-4 text-left text-sm font-semibold text-gray-900 dark:text-white">{{ t('home.comparison.headers.feature') }}</th>
+                  <th class="px-6 py-4 text-center text-sm font-semibold text-gray-500 dark:text-dark-400">{{ t('home.comparison.headers.official') }}</th>
+                  <th class="px-6 py-4 text-center text-sm font-semibold text-primary-600 dark:text-primary-400">{{ t('home.comparison.headers.us') }}</th>
+                </tr>
+              </thead>
+              <tbody class="divide-y divide-gray-200/50 dark:divide-dark-700/50">
+                <tr>
+                  <td class="px-6 py-4 text-sm font-medium text-gray-900 dark:text-white">{{ t('home.comparison.items.pricing.feature') }}</td>
+                  <td class="px-6 py-4 text-center text-sm text-gray-500 dark:text-dark-400">{{ t('home.comparison.items.pricing.official') }}</td>
+                  <td class="px-6 py-4 text-center text-sm font-medium text-primary-600 dark:text-primary-400">{{ t('home.comparison.items.pricing.us') }}</td>
+                </tr>
+                <tr>
+                  <td class="px-6 py-4 text-sm font-medium text-gray-900 dark:text-white">{{ t('home.comparison.items.models.feature') }}</td>
+                  <td class="px-6 py-4 text-center text-sm text-gray-500 dark:text-dark-400">{{ t('home.comparison.items.models.official') }}</td>
+                  <td class="px-6 py-4 text-center text-sm font-medium text-primary-600 dark:text-primary-400">{{ t('home.comparison.items.models.us') }}</td>
+                </tr>
+                <tr>
+                  <td class="px-6 py-4 text-sm font-medium text-gray-900 dark:text-white">{{ t('home.comparison.items.management.feature') }}</td>
+                  <td class="px-6 py-4 text-center text-sm text-gray-500 dark:text-dark-400">{{ t('home.comparison.items.management.official') }}</td>
+                  <td class="px-6 py-4 text-center text-sm font-medium text-primary-600 dark:text-primary-400">{{ t('home.comparison.items.management.us') }}</td>
+                </tr>
+                <tr>
+                  <td class="px-6 py-4 text-sm font-medium text-gray-900 dark:text-white">{{ t('home.comparison.items.stability.feature') }}</td>
+                  <td class="px-6 py-4 text-center text-sm text-gray-500 dark:text-dark-400">{{ t('home.comparison.items.stability.official') }}</td>
+                  <td class="px-6 py-4 text-center text-sm font-medium text-primary-600 dark:text-primary-400">{{ t('home.comparison.items.stability.us') }}</td>
+                </tr>
+                <tr>
+                  <td class="px-6 py-4 text-sm font-medium text-gray-900 dark:text-white">{{ t('home.comparison.items.control.feature') }}</td>
+                  <td class="px-6 py-4 text-center text-sm text-gray-500 dark:text-dark-400">{{ t('home.comparison.items.control.official') }}</td>
+                  <td class="px-6 py-4 text-center text-sm font-medium text-primary-600 dark:text-primary-400">{{ t('home.comparison.items.control.us') }}</td>
+                </tr>
+              </tbody>
+            </table>
+          </div>
+        </div>
+
+        <!-- CTA Section -->
+        <div class="mb-8 rounded-2xl bg-gradient-to-r from-primary-500 to-primary-600 p-8 text-center shadow-xl shadow-primary-500/20 md:p-12">
+          <h2 class="mb-3 text-2xl font-bold text-white md:text-3xl">
+            {{ t('home.cta.title') }}
+          </h2>
+          <p class="mb-6 text-primary-100">
+            {{ t('home.cta.description') }}
+          </p>
+          <router-link
+            v-if="!isAuthenticated"
+            to="/register"
+            class="inline-flex items-center gap-2 rounded-full bg-white px-8 py-3 font-semibold text-primary-600 shadow-lg transition-all hover:bg-gray-50 hover:shadow-xl"
+          >
+            {{ t('home.cta.button') }}
+            <Icon name="arrowRight" size="md" :stroke-width="2" />
+          </router-link>
+          <router-link
+            v-else
+            :to="dashboardPath"
+            class="inline-flex items-center gap-2 rounded-full bg-white px-8 py-3 font-semibold text-primary-600 shadow-lg transition-all hover:bg-gray-50 hover:shadow-xl"
+          >
+            {{ t('home.goToDashboard') }}
+            <Icon name="arrowRight" size="md" :stroke-width="2" />
+          </router-link>
+        </div>
       </div>
     </main>
 
@@ -380,27 +511,20 @@
         <p class="text-sm text-gray-500 dark:text-dark-400">
           &copy; {{ currentYear }} {{ siteName }}. {{ t('home.footer.allRightsReserved') }}
         </p>
-        <div class="flex items-center gap-4">
-          <a
-            v-if="docUrl"
-            :href="docUrl"
-            target="_blank"
-            rel="noopener noreferrer"
-            class="text-sm text-gray-500 transition-colors hover:text-gray-700 dark:text-dark-400 dark:hover:text-white"
-          >
-            {{ t('home.docs') }}
-          </a>
-          <a
-            :href="githubUrl"
-            target="_blank"
-            rel="noopener noreferrer"
-            class="text-sm text-gray-500 transition-colors hover:text-gray-700 dark:text-dark-400 dark:hover:text-white"
-          >
-            GitHub
-          </a>
-        </div>
+        <a
+          v-if="docUrl"
+          :href="docUrl"
+          target="_blank"
+          rel="noopener noreferrer"
+          class="text-sm text-gray-500 transition-colors hover:text-gray-700 dark:text-dark-400 dark:hover:text-white"
+        >
+          {{ t('home.docs') }}
+        </a>
       </div>
     </footer>
+
+    <!-- 微信客服悬浮按钮 -->
+    <WechatServiceButton />
   </div>
 </template>
 
@@ -410,6 +534,7 @@ import { useI18n } from 'vue-i18n'
 import { useAuthStore, useAppStore } from '@/stores'
 import LocaleSwitcher from '@/components/common/LocaleSwitcher.vue'
 import Icon from '@/components/icons/Icon.vue'
+import WechatServiceButton from '@/components/common/WechatServiceButton.vue'
 
 const { t } = useI18n()
 
@@ -419,7 +544,6 @@ const appStore = useAppStore()
 // Site settings - directly from appStore (already initialized from injected config)
 const siteName = computed(() => appStore.cachedPublicSettings?.site_name || appStore.siteName || 'Sub2API')
 const siteLogo = computed(() => appStore.cachedPublicSettings?.site_logo || appStore.siteLogo || '')
-const siteSubtitle = computed(() => appStore.cachedPublicSettings?.site_subtitle || 'AI API Gateway Platform')
 const docUrl = computed(() => appStore.cachedPublicSettings?.doc_url || appStore.docUrl || '')
 const homeContent = computed(() => appStore.cachedPublicSettings?.home_content || '')
 
@@ -432,9 +556,6 @@ const isHomeContentUrl = computed(() => {
 // Theme
 const isDark = ref(document.documentElement.classList.contains('dark'))
 
-// GitHub URL
-const githubUrl = 'https://github.com/Wei-Shaw/sub2api'
-
 // Auth state
 const isAuthenticated = computed(() => authStore.isAuthenticated)
 const isAdmin = computed(() => authStore.isAdmin)

frontend/src/views/admin/GroupsView.vue

@@ -708,6 +708,58 @@
           </div>
         </div>
 
+        <!-- Claude Max Usage 模拟（仅 anthropic 平台） -->
+        <div v-if="createForm.platform === 'anthropic'" class="border-t pt-4">
+          <div class="mb-1.5 flex items-center gap-1">
+            <label class="text-sm font-medium text-gray-700 dark:text-gray-300">
+              {{ t('admin.groups.claudeMaxSimulation.title') }}
+            </label>
+            <div class="group relative inline-flex">
+              <Icon
+                name="questionCircle"
+                size="sm"
+                :stroke-width="2"
+                class="cursor-help text-gray-400 transition-colors hover:text-primary-500 dark:text-gray-500 dark:hover:text-primary-400"
+              />
+              <div class="pointer-events-none absolute bottom-full left-0 z-50 mb-2 w-80 opacity-0 transition-all duration-200 group-hover:pointer-events-auto group-hover:opacity-100">
+                <div class="rounded-lg bg-gray-900 p-3 text-white shadow-lg dark:bg-gray-800">
+                  <p class="text-xs leading-relaxed text-gray-300">
+                    {{ t('admin.groups.claudeMaxSimulation.tooltip') }}
+                  </p>
+                  <div class="absolute -bottom-1.5 left-3 h-3 w-3 rotate-45 bg-gray-900 dark:bg-gray-800"></div>
+                </div>
+              </div>
+            </div>
+          </div>
+          <div class="flex items-center gap-3">
+            <button
+              type="button"
+              @click="createForm.simulate_claude_max_enabled = !createForm.simulate_claude_max_enabled"
+              :class="[
+                'relative inline-flex h-6 w-11 items-center rounded-full transition-colors',
+                createForm.simulate_claude_max_enabled ? 'bg-primary-500' : 'bg-gray-300 dark:bg-dark-600'
+              ]"
+            >
+              <span
+                :class="[
+                  'inline-block h-4 w-4 transform rounded-full bg-white shadow transition-transform',
+                  createForm.simulate_claude_max_enabled ? 'translate-x-6' : 'translate-x-1'
+                ]"
+              />
+            </button>
+            <span class="text-sm text-gray-500 dark:text-gray-400">
+              {{
+                createForm.simulate_claude_max_enabled
+                  ? t('admin.groups.claudeMaxSimulation.enabled')
+                  : t('admin.groups.claudeMaxSimulation.disabled')
+              }}
+            </span>
+          </div>
+          <p class="mt-2 text-xs text-gray-500 dark:text-gray-400">
+            {{ t('admin.groups.claudeMaxSimulation.hint') }}
+          </p>
+        </div>
+
         <!-- 无效请求兜底（仅 anthropic/antigravity 平台，且非订阅分组） -->
         <div
           v-if="['anthropic', 'antigravity'].includes(createForm.platform) && createForm.subscription_type !== 'subscription'"
@@ -1405,6 +1457,58 @@
           </div>
         </div>
 
+        <!-- Claude Max Usage 模拟（仅 anthropic 平台） -->
+        <div v-if="editForm.platform === 'anthropic'" class="border-t pt-4">
+          <div class="mb-1.5 flex items-center gap-1">
+            <label class="text-sm font-medium text-gray-700 dark:text-gray-300">
+              {{ t('admin.groups.claudeMaxSimulation.title') }}
+            </label>
+            <div class="group relative inline-flex">
+              <Icon
+                name="questionCircle"
+                size="sm"
+                :stroke-width="2"
+                class="cursor-help text-gray-400 transition-colors hover:text-primary-500 dark:text-gray-500 dark:hover:text-primary-400"
+              />
+              <div class="pointer-events-none absolute bottom-full left-0 z-50 mb-2 w-80 opacity-0 transition-all duration-200 group-hover:pointer-events-auto group-hover:opacity-100">
+                <div class="rounded-lg bg-gray-900 p-3 text-white shadow-lg dark:bg-gray-800">
+                  <p class="text-xs leading-relaxed text-gray-300">
+                    {{ t('admin.groups.claudeMaxSimulation.tooltip') }}
+                  </p>
+                  <div class="absolute -bottom-1.5 left-3 h-3 w-3 rotate-45 bg-gray-900 dark:bg-gray-800"></div>
+                </div>
+              </div>
+            </div>
+          </div>
+          <div class="flex items-center gap-3">
+            <button
+              type="button"
+              @click="editForm.simulate_claude_max_enabled = !editForm.simulate_claude_max_enabled"
+              :class="[
+                'relative inline-flex h-6 w-11 items-center rounded-full transition-colors',
+                editForm.simulate_claude_max_enabled ? 'bg-primary-500' : 'bg-gray-300 dark:bg-dark-600'
+              ]"
+            >
+              <span
+                :class="[
+                  'inline-block h-4 w-4 transform rounded-full bg-white shadow transition-transform',
+                  editForm.simulate_claude_max_enabled ? 'translate-x-6' : 'translate-x-1'
+                ]"
+              />
+            </button>
+            <span class="text-sm text-gray-500 dark:text-gray-400">
+              {{
+                editForm.simulate_claude_max_enabled
+                  ? t('admin.groups.claudeMaxSimulation.enabled')
+                  : t('admin.groups.claudeMaxSimulation.disabled')
+              }}
+            </span>
+          </div>
+          <p class="mt-2 text-xs text-gray-500 dark:text-gray-400">
+            {{ t('admin.groups.claudeMaxSimulation.hint') }}
+          </p>
+        </div>
+
         <!-- 无效请求兜底（仅 anthropic/antigravity 平台，且非订阅分组） -->
         <div
           v-if="['anthropic', 'antigravity'].includes(editForm.platform) && editForm.subscription_type !== 'subscription'"
@@ -1918,6 +2022,8 @@ const createForm = reactive({
   sora_storage_quota_gb: null as number | null,
   // Claude Code 客户端限制（仅 anthropic 平台使用）
   claude_code_only: false,
+  // Claude Max usage 模拟开关（仅 anthropic 平台）
+  simulate_claude_max_enabled: false,
   fallback_group_id: null as number | null,
   fallback_group_id_on_invalid_request: null as number | null,
   // 模型路由开关
@@ -2159,6 +2265,8 @@ const editForm = reactive({
   sora_storage_quota_gb: null as number | null,
   // Claude Code 客户端限制（仅 anthropic 平台使用）
   claude_code_only: false,
+  // Claude Max usage 模拟开关（仅 anthropic 平台）
+  simulate_claude_max_enabled: false,
   fallback_group_id: null as number | null,
   fallback_group_id_on_invalid_request: null as number | null,
   // 模型路由开关
@@ -2258,6 +2366,7 @@ const closeCreateModal = () => {
   createForm.sora_video_price_per_request_hd = null
   createForm.sora_storage_quota_gb = null
   createForm.claude_code_only = false
+  createForm.simulate_claude_max_enabled = false
   createForm.fallback_group_id = null
   createForm.fallback_group_id_on_invalid_request = null
   createForm.supported_model_scopes = ['claude', 'gemini_text', 'gemini_image']
@@ -2278,6 +2387,8 @@ const handleCreateGroup = async () => {
     const requestData = {
       ...createRest,
       sora_storage_quota_bytes: createQuotaGb ? Math.round(createQuotaGb * 1024 * 1024 * 1024) : 0,
+      simulate_claude_max_enabled:
+        createForm.platform === 'anthropic' ? createForm.simulate_claude_max_enabled : false,
       model_routing: convertRoutingRulesToApiFormat(createModelRoutingRules.value)
     }
     await adminAPI.groups.create(requestData)
@@ -2318,6 +2429,7 @@ const handleEdit = async (group: AdminGroup) => {
   editForm.sora_video_price_per_request_hd = group.sora_video_price_per_request_hd
   editForm.sora_storage_quota_gb = group.sora_storage_quota_bytes ? Number((group.sora_storage_quota_bytes / (1024 * 1024 * 1024)).toFixed(2)) : null
   editForm.claude_code_only = group.claude_code_only || false
+  editForm.simulate_claude_max_enabled = group.simulate_claude_max_enabled || false
   editForm.fallback_group_id = group.fallback_group_id
   editForm.fallback_group_id_on_invalid_request = group.fallback_group_id_on_invalid_request
   editForm.model_routing_enabled = group.model_routing_enabled || false
@@ -2337,6 +2449,7 @@ const closeEditModal = () => {
   showEditModal.value = false
   editingGroup.value = null
   editModelRoutingRules.value = []
+  editForm.simulate_claude_max_enabled = false
   editForm.copy_accounts_from_group_ids = []
 }
 
@@ -2354,6 +2467,8 @@ const handleUpdateGroup = async () => {
     const payload = {
       ...editRest,
       sora_storage_quota_bytes: editQuotaGb ? Math.round(editQuotaGb * 1024 * 1024 * 1024) : 0,
+      simulate_claude_max_enabled:
+        editForm.platform === 'anthropic' ? editForm.simulate_claude_max_enabled : false,
       fallback_group_id: editForm.fallback_group_id === null ? 0 : editForm.fallback_group_id,
       fallback_group_id_on_invalid_request:
         editForm.fallback_group_id_on_invalid_request === null
@@ -2410,6 +2525,21 @@ watch(
     if (!['anthropic', 'antigravity'].includes(newVal)) {
       createForm.fallback_group_id_on_invalid_request = null
     }
+    if (newVal !== 'anthropic') {
+      createForm.simulate_claude_max_enabled = false
+    }
+  }
+)
+
+watch(
+  () => editForm.platform,
+  (newVal) => {
+    if (!['anthropic', 'antigravity'].includes(newVal)) {
+      editForm.fallback_group_id_on_invalid_request = null
+    }
+    if (newVal !== 'anthropic') {
+      editForm.simulate_claude_max_enabled = false
+    }
   }
 )
 

frontend/src/views/admin/ops/components/OpsConcurrencyCard.vue

@@ -122,6 +122,7 @@ const platformRows = computed((): SummaryRow[] => {
       available_accounts: availableAccounts,
       rate_limited_accounts: safeNumber(avail.rate_limit_count),
 
+
       error_accounts: safeNumber(avail.error_count),
       total_concurrency: totalConcurrency,
       used_concurrency: usedConcurrency,
@@ -161,7 +162,6 @@ const groupRows = computed((): SummaryRow[] => {
         total_accounts: totalAccounts,
         available_accounts: availableAccounts,
         rate_limited_accounts: safeNumber(avail.rate_limit_count),
-  
         error_accounts: safeNumber(avail.error_count),
         total_concurrency: totalConcurrency,
         used_concurrency: usedConcurrency,
@@ -329,6 +329,7 @@ function formatDuration(seconds: number): string {
 }
 
 
+
 watch(
   () => realtimeEnabled.value,
   async (enabled) => {

stress_test_gemini_session.sh

@@ -0,0 +1,127 @@
+#!/bin/bash
+
+# Gemini 粘性会话压力测试脚本
+# 测试目标：验证不同会话分配不同账号，同一会话保持同一账号
+
+BASE_URL="http://host.clicodeplus.com:8080"
+API_KEY="sk-32ad0a3197e528c840ea84f0dc6b2056dd3fead03526b5c605a60709bd408f7e"
+MODEL="gemini-2.5-flash"
+
+# 创建临时目录存放结果
+RESULT_DIR="/tmp/gemini_stress_test_$(date +%s)"
+mkdir -p "$RESULT_DIR"
+
+echo "=========================================="
+echo "Gemini 粘性会话压力测试"
+echo "结果目录: $RESULT_DIR"
+echo "=========================================="
+
+# 函数：发送请求并记录
+send_request() {
+    local session_id=$1
+    local round=$2
+    local system_prompt=$3
+    local contents=$4
+    local output_file="$RESULT_DIR/session_${session_id}_round_${round}.json"
+
+    local request_body=$(cat <<EOF
+{
+    "systemInstruction": {
+        "parts": [{"text": "$system_prompt"}]
+    },
+    "contents": $contents
+}
+EOF
+)
+
+    curl -s -X POST "${BASE_URL}/v1beta/models/${MODEL}:generateContent" \
+        -H "Content-Type: application/json" \
+        -H "x-goog-api-key: ${API_KEY}" \
+        -d "$request_body" > "$output_file" 2>&1
+
+    echo "[Session $session_id Round $round] 完成"
+}
+
+# 会话1：数学计算器（累加序列）
+run_session_1() {
+    local sys_prompt="你是一个数学计算器，只返回计算结果数字，不要任何解释"
+
+    # Round 1: 1+1=?
+    send_request 1 1 "$sys_prompt" '[{"role":"user","parts":[{"text":"1+1=?"}]}]'
+
+    # Round 2: 继续 2+2=?（累加历史）
+    send_request 1 2 "$sys_prompt" '[{"role":"user","parts":[{"text":"1+1=?"}]},{"role":"model","parts":[{"text":"2"}]},{"role":"user","parts":[{"text":"2+2=?"}]}]'
+
+    # Round 3: 继续 3+3=?
+    send_request 1 3 "$sys_prompt" '[{"role":"user","parts":[{"text":"1+1=?"}]},{"role":"model","parts":[{"text":"2"}]},{"role":"user","parts":[{"text":"2+2=?"}]},{"role":"model","parts":[{"text":"4"}]},{"role":"user","parts":[{"text":"3+3=?"}]}]'
+
+    # Round 4: 批量计算 10+10, 20+20, 30+30
+    send_request 1 4 "$sys_prompt" '[{"role":"user","parts":[{"text":"1+1=?"}]},{"role":"model","parts":[{"text":"2"}]},{"role":"user","parts":[{"text":"2+2=?"}]},{"role":"model","parts":[{"text":"4"}]},{"role":"user","parts":[{"text":"3+3=?"}]},{"role":"model","parts":[{"text":"6"}]},{"role":"user","parts":[{"text":"计算: 10+10=? 20+20=? 30+30=?"}]}]'
+}
+
+# 会话2：英文翻译器（不同系统提示词 = 不同会话）
+run_session_2() {
+    local sys_prompt="你是一个英文翻译器，将中文翻译成英文，只返回翻译结果"
+
+    send_request 2 1 "$sys_prompt" '[{"role":"user","parts":[{"text":"你好"}]}]'
+    send_request 2 2 "$sys_prompt" '[{"role":"user","parts":[{"text":"你好"}]},{"role":"model","parts":[{"text":"Hello"}]},{"role":"user","parts":[{"text":"世界"}]}]'
+    send_request 2 3 "$sys_prompt" '[{"role":"user","parts":[{"text":"你好"}]},{"role":"model","parts":[{"text":"Hello"}]},{"role":"user","parts":[{"text":"世界"}]},{"role":"model","parts":[{"text":"World"}]},{"role":"user","parts":[{"text":"早上好"}]}]'
+}
+
+# 会话3：日文翻译器
+run_session_3() {
+    local sys_prompt="你是一个日文翻译器，将中文翻译成日文，只返回翻译结果"
+
+    send_request 3 1 "$sys_prompt" '[{"role":"user","parts":[{"text":"你好"}]}]'
+    send_request 3 2 "$sys_prompt" '[{"role":"user","parts":[{"text":"你好"}]},{"role":"model","parts":[{"text":"こんにちは"}]},{"role":"user","parts":[{"text":"谢谢"}]}]'
+    send_request 3 3 "$sys_prompt" '[{"role":"user","parts":[{"text":"你好"}]},{"role":"model","parts":[{"text":"こんにちは"}]},{"role":"user","parts":[{"text":"谢谢"}]},{"role":"model","parts":[{"text":"ありがとう"}]},{"role":"user","parts":[{"text":"再见"}]}]'
+}
+
+# 会话4：乘法计算器（另一个数学会话，但系统提示词不同）
+run_session_4() {
+    local sys_prompt="你是一个乘法专用计算器，只计算乘法，返回数字结果"
+
+    send_request 4 1 "$sys_prompt" '[{"role":"user","parts":[{"text":"2*3=?"}]}]'
+    send_request 4 2 "$sys_prompt" '[{"role":"user","parts":[{"text":"2*3=?"}]},{"role":"model","parts":[{"text":"6"}]},{"role":"user","parts":[{"text":"4*5=?"}]}]'
+    send_request 4 3 "$sys_prompt" '[{"role":"user","parts":[{"text":"2*3=?"}]},{"role":"model","parts":[{"text":"6"}]},{"role":"user","parts":[{"text":"4*5=?"}]},{"role":"model","parts":[{"text":"20"}]},{"role":"user","parts":[{"text":"计算: 10*10=? 20*20=?"}]}]'
+}
+
+# 会话5：诗人（完全不同的角色）
+run_session_5() {
+    local sys_prompt="你是一位诗人，用简短的诗句回应每个话题，每次只写一句诗"
+
+    send_request 5 1 "$sys_prompt" '[{"role":"user","parts":[{"text":"春天"}]}]'
+    send_request 5 2 "$sys_prompt" '[{"role":"user","parts":[{"text":"春天"}]},{"role":"model","parts":[{"text":"春风拂面花满枝"}]},{"role":"user","parts":[{"text":"夏天"}]}]'
+    send_request 5 3 "$sys_prompt" '[{"role":"user","parts":[{"text":"春天"}]},{"role":"model","parts":[{"text":"春风拂面花满枝"}]},{"role":"user","parts":[{"text":"夏天"}]},{"role":"model","parts":[{"text":"蝉鸣蛙声伴荷香"}]},{"role":"user","parts":[{"text":"秋天"}]}]'
+}
+
+echo ""
+echo "开始并发测试 5 个独立会话..."
+echo ""
+
+# 并发运行所有会话
+run_session_1 &
+run_session_2 &
+run_session_3 &
+run_session_4 &
+run_session_5 &
+
+# 等待所有后台任务完成
+wait
+
+echo ""
+echo "=========================================="
+echo "所有请求完成，结果保存在: $RESULT_DIR"
+echo "=========================================="
+
+# 显示结果摘要
+echo ""
+echo "响应摘要:"
+for f in "$RESULT_DIR"/*.json; do
+    filename=$(basename "$f")
+    response=$(cat "$f" | head -c 200)
+    echo "[$filename]: ${response}..."
+done
+
+echo ""
+echo "请检查服务器日志确认账号分配情况"