feat(usage): add service tier tracking and defaults in usage reporting

- Introduced `service_tier` metadata key to capture client-requested service tiers.
- Updated usage records, context propagation, and plugins to include service tier data.
- Added default handling logic for cases where `service_tier` is absent.
- Implemented tests for `service_tier` extraction, defaults, and updates across components.

internal/redisqueue/plugin.go

@@ -52,6 +52,10 @@ func (p *usageQueuePlugin) HandleUsage(ctx context.Context, record coreusage.Rec
 	if reasoningEffort == "" {
 		reasoningEffort = coreusage.ReasoningEffortFromContext(ctx)
 	}
+	serviceTier := strings.TrimSpace(record.ServiceTier)
+	if serviceTier == "" {
+		serviceTier = coreusage.ServiceTierFromContext(ctx)
+	}
 
 	tokens := tokenStats{
 		InputTokens:         record.Detail.InputTokens,
@@ -97,6 +101,7 @@ func (p *usageQueuePlugin) HandleUsage(ctx context.Context, record coreusage.Rec
 		APIKey:          apiKey,
 		RequestID:       requestID,
 		ReasoningEffort: reasoningEffort,
+		ServiceTier:     serviceTier,
 	})
 	if err != nil {
 		return
@@ -114,6 +119,7 @@ type queuedUsageDetail struct {
 	APIKey          string `json:"api_key"`
 	RequestID       string `json:"request_id"`
 	ReasoningEffort string `json:"reasoning_effort"`
+	ServiceTier     string `json:"service_tier"`
 }
 
 type requestDetail struct {

internal/redisqueue/plugin_test.go

@@ -33,6 +33,7 @@ func TestUsageQueuePluginPayloadIncludesStableFieldsAndSuccess(t *testing.T) {
 			AuthType:        "apikey",
 			Source:          "[email protected]",
 			ReasoningEffort: "medium",
+			ServiceTier:     "priority",
 			RequestedAt:     time.Date(2026, 4, 25, 0, 0, 0, 0, time.UTC),
 			Latency:         1500 * time.Millisecond,
 			Detail: coreusage.Detail{
@@ -53,6 +54,7 @@ func TestUsageQueuePluginPayloadIncludesStableFieldsAndSuccess(t *testing.T) {
 		requireMissingField(t, payload, "user_api_key")
 		requireStringField(t, payload, "request_id", "ctx-request-id")
 		requireStringField(t, payload, "reasoning_effort", "medium")
+		requireStringField(t, payload, "service_tier", "priority")
 		requireHeaderField(t, payload, "response_headers", "X-Upstream-Request-Id", []string{"upstream-req-1"})
 		requireHeaderField(t, payload, "response_headers", "Retry-After", []string{"30"})
 		requireBoolField(t, payload, "failed", false)

internal/runtime/executor/helps/usage_helpers.go

@@ -30,6 +30,7 @@ type UsageReporter struct {
 	apiKey      string
 	source      string
 	reasoning   string
+	serviceTier string
 	requestedAt time.Time
 	ttftMu      sync.RWMutex
 	ttft        time.Duration
@@ -53,6 +54,7 @@ func NewUsageReporter(ctx context.Context, provider, model string, auth *cliprox
 		source:      resolveUsageSource(auth, apiKey),
 		authType:    resolveUsageAuthType(auth),
 		reasoning:   usage.ReasoningEffortFromContext(ctx),
+		serviceTier: usage.ServiceTierFromContext(ctx),
 	}
 	if auth != nil {
 		reporter.authID = auth.ID
@@ -78,6 +80,7 @@ func (r *UsageReporter) SetTranslatedReasoningEffort(payload []byte, format stri
 		return
 	}
 	r.reasoning = thinking.ExtractTranslatedReasoningEffort(payload, format)
+	r.serviceTier = extractServiceTierFromPayload(payload)
 }
 
 func (r *UsageReporter) TrackHTTPClient(client *http.Client) *http.Client {
@@ -239,6 +242,7 @@ func (r *UsageReporter) buildRecordForModel(model string, detail usage.Detail, f
 		AuthIndex:       r.authIndex,
 		AuthType:        r.authType,
 		ReasoningEffort: r.reasoning,
+		ServiceTier:     r.serviceTier,
 		RequestedAt:     r.requestedAt,
 		Latency:         r.latency(),
 		TTFT:            r.ttftDuration(),
@@ -248,6 +252,19 @@ func (r *UsageReporter) buildRecordForModel(model string, detail usage.Detail, f
 	}
 }
 
+func extractServiceTierFromPayload(payload []byte) string {
+	if len(payload) == 0 {
+		return usage.DefaultServiceTier
+	}
+	for _, path := range []string{"service_tier", "request.service_tier", "response.service_tier"} {
+		serviceTier := strings.TrimSpace(gjson.GetBytes(payload, path).String())
+		if serviceTier != "" {
+			return serviceTier
+		}
+	}
+	return usage.DefaultServiceTier
+}
+
 func failFromErrors(errs ...error) usage.Failure {
 	for _, err := range errs {
 		if err == nil {

internal/runtime/executor/helps/usage_helpers_test.go

@@ -241,6 +241,39 @@ func TestUsageReporterBuildRecordIncludesReasoningEffort(t *testing.T) {
 	}
 }
 
+func TestUsageReporterBuildRecordIncludesServiceTier(t *testing.T) {
+	ctx := usage.WithServiceTier(context.Background(), "priority")
+	reporter := NewUsageReporter(ctx, "openai", "gpt-5.4", nil)
+
+	record := reporter.buildRecord(usage.Detail{TotalTokens: 3}, false)
+	if record.ServiceTier != "priority" {
+		t.Fatalf("service tier = %q, want %q", record.ServiceTier, "priority")
+	}
+}
+
+func TestUsageReporterSetTranslatedReasoningEffortUpdatesServiceTier(t *testing.T) {
+	reporter := NewUsageReporter(context.Background(), "openai", "gpt-5.4", nil)
+
+	reporter.SetTranslatedReasoningEffort([]byte(`{"service_tier":"priority"}`), "openai")
+
+	record := reporter.buildRecord(usage.Detail{TotalTokens: 3}, false)
+	if record.ServiceTier != "priority" {
+		t.Fatalf("service tier = %q, want %q", record.ServiceTier, "priority")
+	}
+}
+
+func TestUsageReporterSetTranslatedReasoningEffortDefaultsServiceTierWhenRemoved(t *testing.T) {
+	ctx := usage.WithServiceTier(context.Background(), "priority")
+	reporter := NewUsageReporter(ctx, "openai", "gpt-5.4", nil)
+
+	reporter.SetTranslatedReasoningEffort([]byte(`{"model":"gpt-5.4"}`), "openai")
+
+	record := reporter.buildRecord(usage.Detail{TotalTokens: 3}, false)
+	if record.ServiceTier != usage.DefaultServiceTier {
+		t.Fatalf("service tier = %q, want %q", record.ServiceTier, usage.DefaultServiceTier)
+	}
+}
+
 func TestUsageReporterBuildAdditionalModelRecordSkipsZeroTokens(t *testing.T) {
 	reporter := &UsageReporter{
 		provider:    "codex",

sdk/api/handlers/handlers.go

@@ -20,8 +20,10 @@ import (
 	"github.com/router-for-me/CLIProxyAPI/v7/internal/util"
 	coreauth "github.com/router-for-me/CLIProxyAPI/v7/sdk/cliproxy/auth"
 	coreexecutor "github.com/router-for-me/CLIProxyAPI/v7/sdk/cliproxy/executor"
+	coreusage "github.com/router-for-me/CLIProxyAPI/v7/sdk/cliproxy/usage"
 	"github.com/router-for-me/CLIProxyAPI/v7/sdk/config"
 	sdktranslator "github.com/router-for-me/CLIProxyAPI/v7/sdk/translator"
+	"github.com/tidwall/gjson"
 	"golang.org/x/net/context"
 )
 
@@ -242,6 +244,21 @@ func setReasoningEffortMetadata(meta map[string]any, handlerType, model string,
 	meta[coreexecutor.ReasoningEffortMetadataKey] = effort
 }
 
+func setServiceTierMetadata(meta map[string]any, rawJSON []byte) {
+	if meta == nil {
+		return
+	}
+	serviceTier := coreusage.DefaultServiceTier
+	node := gjson.GetBytes(rawJSON, "service_tier")
+	if node.Exists() {
+		value := strings.TrimSpace(node.String())
+		if value != "" {
+			serviceTier = value
+		}
+	}
+	meta[coreexecutor.ServiceTierMetadataKey] = serviceTier
+}
+
 // headersFromContext extracts the original HTTP request headers from the gin context
 // embedded in the provided context. This allows session affinity selectors to read
 // client headers like X-Amp-Thread-Id.
@@ -562,6 +579,7 @@ func (h *BaseAPIHandler) executeWithAuthManager(ctx context.Context, handlerType
 	reqMeta := requestExecutionMetadata(ctx)
 	reqMeta[coreexecutor.RequestedModelMetadataKey] = modelName
 	setReasoningEffortMetadata(reqMeta, handlerType, normalizedModel, rawJSON)
+	setServiceTierMetadata(reqMeta, rawJSON)
 	payload := rawJSON
 	if len(payload) == 0 {
 		payload = nil
@@ -611,6 +629,7 @@ func (h *BaseAPIHandler) ExecuteCountWithAuthManager(ctx context.Context, handle
 	reqMeta := requestExecutionMetadata(ctx)
 	reqMeta[coreexecutor.RequestedModelMetadataKey] = modelName
 	setReasoningEffortMetadata(reqMeta, handlerType, normalizedModel, rawJSON)
+	setServiceTierMetadata(reqMeta, rawJSON)
 	payload := rawJSON
 	if len(payload) == 0 {
 		payload = nil
@@ -673,6 +692,7 @@ func (h *BaseAPIHandler) executeStreamWithAuthManager(ctx context.Context, handl
 	reqMeta := requestExecutionMetadata(ctx)
 	reqMeta[coreexecutor.RequestedModelMetadataKey] = modelName
 	setReasoningEffortMetadata(reqMeta, handlerType, normalizedModel, rawJSON)
+	setServiceTierMetadata(reqMeta, rawJSON)
 	payload := rawJSON
 	if len(payload) == 0 {
 		payload = nil

sdk/api/handlers/handlers_metadata_test.go

@@ -38,3 +38,25 @@ func TestSetReasoningEffortMetadataSupportsOpenAIResponses(t *testing.T) {
 		t.Fatalf("ReasoningEffortMetadataKey = %v, want %q", got, "medium")
 	}
 }
+
+func TestSetServiceTierMetadataExtractsValue(t *testing.T) {
+	meta := make(map[string]any)
+
+	setServiceTierMetadata(meta, []byte(`{"service_tier":"priority"}`))
+
+	gotServiceTier := meta[coreexecutor.ServiceTierMetadataKey]
+	if gotServiceTier != "priority" {
+		t.Fatalf("ServiceTierMetadataKey = %v, want %q", gotServiceTier, "priority")
+	}
+}
+
+func TestSetServiceTierMetadataDefaultsWhenMissing(t *testing.T) {
+	meta := make(map[string]any)
+
+	setServiceTierMetadata(meta, []byte(`{"model":"gpt-5.4"}`))
+
+	gotServiceTier := meta[coreexecutor.ServiceTierMetadataKey]
+	if gotServiceTier != "default" {
+		t.Fatalf("ServiceTierMetadataKey = %v, want %q", gotServiceTier, "default")
+	}
+}

sdk/cliproxy/auth/conductor.go

@@ -1731,9 +1731,14 @@ func (m *Manager) prepareRequestAuth(ctx context.Context, executor ProviderExecu
 func contextWithRequestedModelAlias(ctx context.Context, opts cliproxyexecutor.Options, fallback string) context.Context {
 	alias := requestedModelAliasFromOptions(opts, fallback)
 	ctx = coreusage.WithRequestedModelAlias(ctx, alias)
-	if effort := reasoningEffortFromOptions(opts); effort != "" {
+	effort := reasoningEffortFromOptions(opts)
+	if effort != "" {
 		ctx = coreusage.WithReasoningEffort(ctx, effort)
 	}
+	serviceTier := serviceTierFromOptions(opts)
+	if serviceTier != "" {
+		ctx = coreusage.WithServiceTier(ctx, serviceTier)
+	}
 	return ctx
 }
 
@@ -1780,6 +1785,24 @@ func reasoningEffortFromOptions(opts cliproxyexecutor.Options) string {
 	}
 }
 
+func serviceTierFromOptions(opts cliproxyexecutor.Options) string {
+	if len(opts.Metadata) == 0 {
+		return ""
+	}
+	raw, ok := opts.Metadata[cliproxyexecutor.ServiceTierMetadataKey]
+	if !ok || raw == nil {
+		return ""
+	}
+	switch value := raw.(type) {
+	case string:
+		return strings.TrimSpace(value)
+	case []byte:
+		return strings.TrimSpace(string(value))
+	default:
+		return ""
+	}
+}
+
 func pinnedAuthIDFromMetadata(meta map[string]any) string {
 	if len(meta) == 0 {
 		return ""

sdk/cliproxy/auth/conductor_usage_test.go

@@ -13,6 +13,7 @@ func TestContextWithRequestedModelAliasIncludesReasoningEffort(t *testing.T) {
 		Metadata: map[string]any{
 			cliproxyexecutor.RequestedModelMetadataKey:  "client-model",
 			cliproxyexecutor.ReasoningEffortMetadataKey: "medium",
+			cliproxyexecutor.ServiceTierMetadataKey:     "priority",
 		},
 	}, "fallback-model")
 
@@ -22,4 +23,8 @@ func TestContextWithRequestedModelAliasIncludesReasoningEffort(t *testing.T) {
 	if got := coreusage.ReasoningEffortFromContext(ctx); got != "medium" {
 		t.Fatalf("reasoning effort = %q, want %q", got, "medium")
 	}
+	gotServiceTier := coreusage.ServiceTierFromContext(ctx)
+	if gotServiceTier != "priority" {
+		t.Fatalf("service tier = %q, want %q", gotServiceTier, "priority")
+	}
 }

sdk/cliproxy/executor/types.go

@@ -20,6 +20,9 @@ const DisallowFreeAuthMetadataKey = "disallow_free_auth"
 // ReasoningEffortMetadataKey stores the client-requested reasoning effort for usage logs.
 const ReasoningEffortMetadataKey = "reasoning_effort"
 
+// ServiceTierMetadataKey stores the client-requested service tier for usage logs.
+const ServiceTierMetadataKey = "service_tier"
+
 const (
 	// PinnedAuthMetadataKey locks execution to a specific auth ID.
 	PinnedAuthMetadataKey = "pinned_auth_id"

sdk/cliproxy/usage/manager.go

@@ -10,6 +10,9 @@ import (
 	log "github.com/sirupsen/logrus"
 )
 
+// DefaultServiceTier is used when a request does not specify service_tier.
+const DefaultServiceTier = "default"
+
 // Record contains the usage statistics captured for a single provider request.
 type Record struct {
 	Provider  string
@@ -22,12 +25,14 @@ type Record struct {
 	Source    string
 	// ReasoningEffort stores the translated upstream thinking level for request event logs.
 	ReasoningEffort string
-	RequestedAt     time.Time
-	Latency         time.Duration
-	TTFT            time.Duration
-	Failed          bool
-	Fail            Failure
-	Detail          Detail
+	// ServiceTier stores the client-requested service tier for request event logs.
+	ServiceTier string
+	RequestedAt time.Time
+	Latency     time.Duration
+	TTFT        time.Duration
+	Failed      bool
+	Fail        Failure
+	Detail      Detail
 	// ResponseHeaders stores a snapshot of upstream response headers for usage sinks.
 	ResponseHeaders http.Header
 }
@@ -51,6 +56,7 @@ type Detail struct {
 
 type requestedModelAliasContextKey struct{}
 type reasoningEffortContextKey struct{}
+type serviceTierContextKey struct{}
 
 // WithRequestedModelAlias stores the client-requested model name for usage sinks.
 func WithRequestedModelAlias(ctx context.Context, alias string) context.Context {
@@ -108,6 +114,42 @@ func ReasoningEffortFromContext(ctx context.Context) string {
 	}
 }
 
+// WithServiceTier stores the client-requested service tier for usage sinks.
+func WithServiceTier(ctx context.Context, tier string) context.Context {
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	tier = strings.TrimSpace(tier)
+	if tier == "" {
+		tier = DefaultServiceTier
+	}
+	return context.WithValue(ctx, serviceTierContextKey{}, tier)
+}
+
+// ServiceTierFromContext returns the client-requested service tier stored in ctx.
+func ServiceTierFromContext(ctx context.Context) string {
+	if ctx == nil {
+		return DefaultServiceTier
+	}
+	raw := ctx.Value(serviceTierContextKey{})
+	switch value := raw.(type) {
+	case string:
+		tier := strings.TrimSpace(value)
+		if tier == "" {
+			return DefaultServiceTier
+		}
+		return tier
+	case []byte:
+		tier := strings.TrimSpace(string(value))
+		if tier == "" {
+			return DefaultServiceTier
+		}
+		return tier
+	default:
+		return DefaultServiceTier
+	}
+}
+
 // Plugin consumes usage records emitted by the proxy runtime.
 type Plugin interface {
 	HandleUsage(ctx context.Context, record Record)