feat(usage): report request and response service tiers

internal/redisqueue/plugin.go

@@ -56,10 +56,14 @@ 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)
+	requestServiceTier := strings.TrimSpace(record.RequestServiceTier)
+	if requestServiceTier == "" {
+		requestServiceTier = strings.TrimSpace(record.ServiceTier)
 	}
+	if requestServiceTier == "" {
+		requestServiceTier = coreusage.ServiceTierFromContext(ctx)
+	}
+	responseServiceTier := strings.TrimSpace(record.ResponseServiceTier)
 
 	tokens := tokenStats{
 		InputTokens:         record.Detail.InputTokens,
@@ -96,17 +100,19 @@ func (p *usageQueuePlugin) HandleUsage(ctx context.Context, record coreusage.Rec
 	}
 
 	payload, err := json.Marshal(queuedUsageDetail{
-		requestDetail:   detail,
-		Provider:        provider,
-		ExecutorType:    executorType,
-		Model:           modelName,
-		Alias:           aliasName,
-		Endpoint:        resolveEndpoint(ctx),
-		AuthType:        authType,
-		APIKey:          apiKey,
-		RequestID:       requestID,
-		ReasoningEffort: reasoningEffort,
-		ServiceTier:     serviceTier,
+		requestDetail:       detail,
+		Provider:            provider,
+		ExecutorType:        executorType,
+		Model:               modelName,
+		Alias:               aliasName,
+		Endpoint:            resolveEndpoint(ctx),
+		AuthType:            authType,
+		APIKey:              apiKey,
+		RequestID:           requestID,
+		ReasoningEffort:     reasoningEffort,
+		ServiceTier:         requestServiceTier,
+		RequestServiceTier:  requestServiceTier,
+		ResponseServiceTier: responseServiceTier,
 	})
 	if err != nil {
 		return
@@ -116,16 +122,18 @@ func (p *usageQueuePlugin) HandleUsage(ctx context.Context, record coreusage.Rec
 
 type queuedUsageDetail struct {
 	requestDetail
-	Provider        string `json:"provider"`
-	ExecutorType    string `json:"executor_type"`
-	Model           string `json:"model"`
-	Alias           string `json:"alias"`
-	Endpoint        string `json:"endpoint"`
-	AuthType        string `json:"auth_type"`
-	APIKey          string `json:"api_key"`
-	RequestID       string `json:"request_id"`
-	ReasoningEffort string `json:"reasoning_effort"`
-	ServiceTier     string `json:"service_tier"`
+	Provider            string `json:"provider"`
+	ExecutorType        string `json:"executor_type"`
+	Model               string `json:"model"`
+	Alias               string `json:"alias"`
+	Endpoint            string `json:"endpoint"`
+	AuthType            string `json:"auth_type"`
+	APIKey              string `json:"api_key"`
+	RequestID           string `json:"request_id"`
+	ReasoningEffort     string `json:"reasoning_effort"`
+	ServiceTier         string `json:"service_tier"`
+	RequestServiceTier  string `json:"request_service_tier"`
+	ResponseServiceTier string `json:"response_service_tier,omitempty"`
 }
 
 type requestDetail struct {

internal/redisqueue/plugin_test.go

@@ -25,18 +25,19 @@ func TestUsageQueuePluginPayloadIncludesStableFieldsAndSuccess(t *testing.T) {
 
 		plugin := &usageQueuePlugin{}
 		plugin.HandleUsage(ctx, coreusage.Record{
-			Provider:        "openai",
-			ExecutorType:    "KimiExecutor",
-			Model:           "gpt-5.4",
-			Alias:           "client-gpt",
-			APIKey:          "test-key",
-			AuthIndex:       "0",
-			AuthType:        "apikey",
-			Source:          "user@example.com",
-			ReasoningEffort: "medium",
-			ServiceTier:     "priority",
-			RequestedAt:     time.Date(2026, 4, 25, 0, 0, 0, 0, time.UTC),
-			Latency:         1500 * time.Millisecond,
+			Provider:            "openai",
+			ExecutorType:        "KimiExecutor",
+			Model:               "gpt-5.4",
+			Alias:               "client-gpt",
+			APIKey:              "test-key",
+			AuthIndex:           "0",
+			AuthType:            "apikey",
+			Source:              "user@example.com",
+			ReasoningEffort:     "medium",
+			ServiceTier:         "priority",
+			ResponseServiceTier: "default",
+			RequestedAt:         time.Date(2026, 4, 25, 0, 0, 0, 0, time.UTC),
+			Latency:             1500 * time.Millisecond,
 			Detail: coreusage.Detail{
 				InputTokens:  10,
 				OutputTokens: 20,
@@ -57,6 +58,8 @@ func TestUsageQueuePluginPayloadIncludesStableFieldsAndSuccess(t *testing.T) {
 		requireStringField(t, payload, "request_id", "ctx-request-id")
 		requireStringField(t, payload, "reasoning_effort", "medium")
 		requireStringField(t, payload, "service_tier", "priority")
+		requireStringField(t, payload, "request_service_tier", "priority")
+		requireStringField(t, payload, "response_service_tier", "default")
 		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

@@ -260,23 +260,25 @@ func (r *UsageReporter) buildRecordForModel(model string, detail usage.Detail, f
 		return usage.Record{Model: model, Detail: detail, Failed: failed, Fail: fail}
 	}
 	return usage.Record{
-		Provider:        r.provider,
-		ExecutorType:    r.executorType,
-		Model:           model,
-		Alias:           r.alias,
-		Source:          r.source,
-		APIKey:          r.apiKey,
-		AuthID:          r.authID,
-		AuthIndex:       r.authIndex,
-		AuthType:        r.authType,
-		ReasoningEffort: r.reasoning,
-		ServiceTier:     r.serviceTier,
-		RequestedAt:     r.requestedAt,
-		Latency:         r.latency(),
-		TTFT:            r.ttftDuration(),
-		Failed:          failed,
-		Fail:            fail,
-		Detail:          detail,
+		Provider:            r.provider,
+		ExecutorType:        r.executorType,
+		Model:               model,
+		Alias:               r.alias,
+		Source:              r.source,
+		APIKey:              r.apiKey,
+		AuthID:              r.authID,
+		AuthIndex:           r.authIndex,
+		AuthType:            r.authType,
+		ReasoningEffort:     r.reasoning,
+		ServiceTier:         r.serviceTier,
+		RequestServiceTier:  r.serviceTier,
+		ResponseServiceTier: strings.TrimSpace(detail.ResponseServiceTier),
+		RequestedAt:         r.requestedAt,
+		Latency:             r.latency(),
+		TTFT:                r.ttftDuration(),
+		Failed:              failed,
+		Fail:                fail,
+		Detail:              detail,
 	}
 }
 
@@ -484,7 +486,9 @@ func ParseCodexUsage(data []byte) (usage.Detail, bool) {
 	if !hasOpenAIStyleUsageTokenFields(usageNode) {
 		return usage.Detail{}, false
 	}
-	return parseOpenAIStyleUsageNode(usageNode), true
+	detail := parseOpenAIStyleUsageNode(usageNode)
+	detail.ResponseServiceTier = extractResponseServiceTier(data)
+	return detail, true
 }
 
 func ParseCodexImageToolUsage(data []byte) (usage.Detail, bool) {
@@ -500,7 +504,9 @@ func ParseOpenAIUsage(data []byte) usage.Detail {
 	if !hasOpenAIStyleUsageTokenFields(usageNode) {
 		return usage.Detail{}
 	}
-	return parseOpenAIStyleUsageNode(usageNode)
+	detail := parseOpenAIStyleUsageNode(usageNode)
+	detail.ResponseServiceTier = extractResponseServiceTier(data)
+	return detail
 }
 
 func hasOpenAIStyleUsageTokenFields(usageNode gjson.Result) bool {
@@ -562,7 +568,9 @@ func ParseOpenAIStreamUsage(line []byte) (usage.Detail, bool) {
 	if !hasOpenAIStyleUsageTokenFields(usageNode) {
 		return usage.Detail{}, false
 	}
-	return parseOpenAIStyleUsageNode(usageNode), true
+	detail := parseOpenAIStyleUsageNode(usageNode)
+	detail.ResponseServiceTier = extractResponseServiceTier(payload)
+	return detail, true
 }
 
 func ParseClaudeUsage(data []byte) usage.Detail {
@@ -643,9 +651,25 @@ func ParseInteractionsUsage(data []byte) usage.Detail {
 		return usage.Detail{}
 	}
 	if node.Get("promptTokenCount").Exists() || node.Get("candidatesTokenCount").Exists() {
-		return parseGeminiFamilyUsageDetail(node)
+		detail := parseGeminiFamilyUsageDetail(node)
+		detail.ResponseServiceTier = extractResponseServiceTier(data)
+		return detail
 	}
-	return parseInteractionsUsageDetail(node)
+	detail := parseInteractionsUsageDetail(node)
+	detail.ResponseServiceTier = extractResponseServiceTier(data)
+	return detail
+}
+
+func extractResponseServiceTier(payload []byte) string {
+	if len(payload) == 0 || !gjson.ValidBytes(payload) {
+		return ""
+	}
+	for _, path := range []string{"response.service_tier", "service_tier", "interaction.service_tier"} {
+		if tier := strings.TrimSpace(gjson.GetBytes(payload, path).String()); tier != "" {
+			return tier
+		}
+	}
+	return ""
 }
 
 func ParseInteractionsStreamUsage(line []byte) (usage.Detail, bool) {

internal/runtime/executor/helps/usage_helpers_test.go

@@ -32,7 +32,7 @@ func TestParseOpenAIUsageChatCompletions(t *testing.T) {
 }
 
 func TestParseOpenAIUsageResponses(t *testing.T) {
-	data := []byte(`{"usage":{"input_tokens":10,"output_tokens":20,"total_tokens":30,"input_tokens_details":{"cached_tokens":7},"output_tokens_details":{"reasoning_tokens":9}}}`)
+	data := []byte(`{"service_tier":"default","usage":{"input_tokens":10,"output_tokens":20,"total_tokens":30,"input_tokens_details":{"cached_tokens":7},"output_tokens_details":{"reasoning_tokens":9}}}`)
 	detail := ParseOpenAIUsage(data)
 	if detail.InputTokens != 10 {
 		t.Fatalf("input tokens = %d, want %d", detail.InputTokens, 10)
@@ -49,10 +49,13 @@ func TestParseOpenAIUsageResponses(t *testing.T) {
 	if detail.ReasoningTokens != 9 {
 		t.Fatalf("reasoning tokens = %d, want %d", detail.ReasoningTokens, 9)
 	}
+	if detail.ResponseServiceTier != "default" {
+		t.Fatalf("response service tier = %q, want default", detail.ResponseServiceTier)
+	}
 }
 
 func TestParseCodexUsageIncludesCacheWriteTokens(t *testing.T) {
-	data := []byte(`{"response":{"usage":{"input_tokens":100,"output_tokens":20,"total_tokens":120,"input_tokens_details":{"cached_tokens":30,"cache_write_tokens":40}}}}`)
+	data := []byte(`{"response":{"service_tier":"priority","usage":{"input_tokens":100,"output_tokens":20,"total_tokens":120,"input_tokens_details":{"cached_tokens":30,"cache_write_tokens":40}}}}`)
 	detail, ok := ParseCodexUsage(data)
 	if !ok {
 		t.Fatal("ParseCodexUsage() ok = false, want true")
@@ -72,6 +75,9 @@ func TestParseCodexUsageIncludesCacheWriteTokens(t *testing.T) {
 	if detail.TotalTokens != 120 {
 		t.Fatalf("total tokens = %d, want 120", detail.TotalTokens)
 	}
+	if detail.ResponseServiceTier != "priority" {
+		t.Fatalf("response service tier = %q, want priority", detail.ResponseServiceTier)
+	}
 }
 
 func TestParseOpenAIUsageIgnoresNullUsage(t *testing.T) {
@@ -90,7 +96,7 @@ func TestParseOpenAIStreamUsageIgnoresNullUsage(t *testing.T) {
 }
 
 func TestParseOpenAIStreamUsageResponsesFields(t *testing.T) {
-	line := []byte(`data: {"id":"chunk_1","object":"chat.completion.chunk","choices":[],"usage":{"input_tokens":8,"output_tokens":5,"total_tokens":13,"input_tokens_details":{"cached_tokens":3},"output_tokens_details":{"reasoning_tokens":2}}}`)
+	line := []byte(`data: {"id":"chunk_1","object":"chat.completion.chunk","service_tier":"flex","choices":[],"usage":{"input_tokens":8,"output_tokens":5,"total_tokens":13,"input_tokens_details":{"cached_tokens":3},"output_tokens_details":{"reasoning_tokens":2}}}`)
 	detail, ok := ParseOpenAIStreamUsage(line)
 	if !ok {
 		t.Fatal("ParseOpenAIStreamUsage() ok = false, want true")
@@ -110,6 +116,9 @@ func TestParseOpenAIStreamUsageResponsesFields(t *testing.T) {
 	if detail.ReasoningTokens != 2 {
 		t.Fatalf("reasoning tokens = %d, want %d", detail.ReasoningTokens, 2)
 	}
+	if detail.ResponseServiceTier != "flex" {
+		t.Fatalf("response service tier = %q, want flex", detail.ResponseServiceTier)
+	}
 }
 
 func TestStreamUsageBufferKeepsLastUsage(t *testing.T) {
@@ -324,10 +333,16 @@ 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)
+	record := reporter.buildRecord(usage.Detail{TotalTokens: 3, ResponseServiceTier: "default"}, false)
 	if record.ServiceTier != "priority" {
 		t.Fatalf("service tier = %q, want %q", record.ServiceTier, "priority")
 	}
+	if record.RequestServiceTier != "priority" {
+		t.Fatalf("request service tier = %q, want priority", record.RequestServiceTier)
+	}
+	if record.ResponseServiceTier != "default" {
+		t.Fatalf("response service tier = %q, want default", record.ResponseServiceTier)
+	}
 }
 
 func TestUsageReporterSetTranslatedReasoningEffortUpdatesServiceTier(t *testing.T) {

sdk/cliproxy/usage/manager.go

@@ -29,12 +29,16 @@ type Record struct {
 	ReasoningEffort string
 	// 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
+	// RequestServiceTier explicitly aliases the client-requested service tier.
+	RequestServiceTier string
+	// ResponseServiceTier stores the final tier reported by the upstream response.
+	ResponseServiceTier 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
 }
@@ -54,6 +58,7 @@ type Detail struct {
 	CacheReadTokens     int64
 	CacheCreationTokens int64
 	TotalTokens         int64
+	ResponseServiceTier string
 }
 
 type requestedModelAliasContextKey struct{}