Merge pull request #4875 from StarryKira/codex/fix-4859-gemini-36-flash-billing

fix(billing): price Antigravity Gemini 3.6 Flash

backend/internal/service/account_stats_pricing_test.go

@@ -7,6 +7,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/Wei-Shaw/sub2api/internal/config"
 	"github.com/stretchr/testify/require"
 )
 
@@ -698,6 +699,25 @@ func TestResolveAccountStatsCost_FallsBackToLiteLLM(t *testing.T) {
 	require.InDelta(t, 0.2, *result, 1e-12)
 }
 
+func TestResolveAccountStatsCost_Gemini36FlashTierUsesFallbackPricing(t *testing.T) {
+	channel := &Channel{
+		ID:                         1,
+		Status:                     StatusActive,
+		ApplyPricingToAccountStats: false,
+	}
+	cs := newTestChannelServiceForStats(t, channel, 10, "antigravity")
+	bs := NewBillingService(&config.Config{}, nil)
+
+	result := resolveAccountStatsCost(
+		context.Background(),
+		cs, bs,
+		1, 10, "gemini-3.6-flash-low",
+		UsageTokens{InputTokens: 1_000_000, OutputTokens: 1_000_000, CacheReadTokens: 1_000_000}, 1, 0,
+	)
+	require.NotNil(t, result)
+	require.InDelta(t, 9.15, *result, 1e-12)
+}
+
 func TestResolveAccountStatsCost_AllMiss_ReturnsNil(t *testing.T) {
 	channel := &Channel{
 		ID:                         1,

backend/internal/service/billing_service.go

@@ -272,6 +272,17 @@ func (s *BillingService) initFallbackPricing() {
 		SupportsCacheBreakdown:     false,
 	}
 
+	// Gemini 3.6 Flash (Google AI pricing: $1.50 input / $7.50 output /
+	// $0.15 cached input per MTok). Antigravity's -high/-low/-medium/-tiered
+	// aliases are matched below so unavailable remote pricing never records
+	// token-bearing requests at $0.
+	s.fallbackPrices["gemini-3.6-flash"] = &ModelPricing{
+		InputPricePerToken:     1.5e-6,
+		OutputPricePerToken:    7.5e-6,
+		CacheReadPricePerToken: 0.15e-6,
+		SupportsCacheBreakdown: false,
+	}
+
 	// OpenAI GPT-5.4（业务指定价格）
 	s.fallbackPrices["gpt-5.4"] = &ModelPricing{
 		InputPricePerToken:             2.5e-6,  // $2.5 per MTok
@@ -630,6 +641,9 @@ func (s *BillingService) getFallbackPricing(model string) *ModelPricing {
 	if strings.Contains(modelLower, "gemini-3.1-pro") || strings.Contains(modelLower, "gemini-3-1-pro") {
 		return s.fallbackPrices["gemini-3.1-pro"]
 	}
+	if strings.Contains(modelLower, "gemini-3.6-flash") || strings.Contains(modelLower, "gemini-3-6-flash") {
+		return s.fallbackPrices["gemini-3.6-flash"]
+	}
 
 	// DeepSeek V4 系列：仅匹配已知 V4 Pro/Flash 与官方兼容别名
 	// （deepseek-chat / deepseek-reasoner → V4 Flash），未知 deepseek-* 型号不回退，避免误计价。

backend/internal/service/pricing_service.go

@@ -695,16 +695,24 @@ func (s *PricingService) GetModelPricing(modelName string) *LiteLLMModelPricing
 }
 
 func (s *PricingService) buildModelLookupCandidates(modelLower string) []string {
-	// Prefer canonical model name first (this also improves billing compatibility with "models/xxx").
-	candidates := []string{
-		normalizeModelNameForPricing(modelLower),
+	rawCandidates := []string{
 		modelLower,
-	}
-	candidates = append(candidates,
 		strings.TrimPrefix(modelLower, "models/"),
 		lastSegment(modelLower),
 		lastSegment(strings.TrimPrefix(modelLower, "models/")),
-	)
+	}
+	normalized := normalizeModelNameForPricing(modelLower)
+
+	// A tier-specific entry should take precedence when the pricing catalog gains
+	// one later. Today Antigravity's Gemini 3.6 Flash tiers share the base rate,
+	// so the normalized base remains the fallback after the exact aliases.
+	candidates := rawCandidates
+	if normalizeGeminiThinkingTierAlias(lastSegment(modelLower)) != lastSegment(modelLower) {
+		candidates = append(candidates, normalized)
+	} else {
+		// Prefer canonical model names for all other aliases (including models/xxx).
+		candidates = append([]string{normalized}, candidates...)
+	}
 
 	seen := make(map[string]struct{}, len(candidates))
 	out := make([]string, 0, len(candidates))
@@ -752,6 +760,20 @@ func normalizeModelNameForPricing(model string) string {
 		}
 		return canonical
 	}
+	return normalizeGeminiThinkingTierAlias(model)
+}
+
+// normalizeGeminiThinkingTierAlias maps Antigravity's Gemini 3.6 Flash
+// thinking-tier model IDs to the public base model. The tier controls reasoning
+// behavior, not the published token rate, so this keeps -high/-low/-medium and
+// -tiered requests on the same price card as gemini-3.6-flash.
+func normalizeGeminiThinkingTierAlias(model string) string {
+	const baseModel = "gemini-3.6-flash"
+	for _, tier := range []string{"-high", "-low", "-medium", "-tiered"} {
+		if model == baseModel+tier {
+			return baseModel
+		}
+	}
 	return model
 }
 

backend/internal/service/pricing_service_test.go

@@ -454,6 +454,83 @@ func TestGetModelPricing_OpenAICompactAliasUsesStaticFallback(t *testing.T) {
 	require.InDelta(t, 1.5e-5, got.OutputCostPerToken, 1e-12)
 }
 
+func TestPricingService_Gemini36FlashThinkingTiersUseBasePricing(t *testing.T) {
+	basePricing := &LiteLLMModelPricing{
+		InputCostPerToken:       1.5e-6,
+		OutputCostPerToken:      7.5e-6,
+		CacheReadInputTokenCost: 0.15e-6,
+	}
+	svc := &PricingService{pricingData: map[string]*LiteLLMModelPricing{
+		"gemini-3.6-flash": basePricing,
+	}}
+
+	for _, model := range []string{
+		"gemini-3.6-flash",
+		"gemini-3.6-flash-high",
+		"gemini-3.6-flash-low",
+		"gemini-3.6-flash-medium",
+		"gemini-3.6-flash-tiered",
+	} {
+		t.Run(model, func(t *testing.T) {
+			require.Same(t, basePricing, svc.GetModelPricing(model))
+		})
+	}
+}
+
+func TestPricingService_Gemini36FlashTierSpecificPricingTakesPrecedence(t *testing.T) {
+	basePricing := &LiteLLMModelPricing{InputCostPerToken: 1.5e-6}
+	tierPricing := &LiteLLMModelPricing{InputCostPerToken: 2e-6}
+	svc := &PricingService{pricingData: map[string]*LiteLLMModelPricing{
+		"gemini-3.6-flash":     basePricing,
+		"gemini-3.6-flash-low": tierPricing,
+	}}
+
+	require.Same(t, tierPricing, svc.GetModelPricing("models/gemini-3.6-flash-low"))
+}
+
+func TestBillingService_Gemini36FlashThinkingTierFallbacksAreBillable(t *testing.T) {
+	svc := NewBillingService(&config.Config{}, nil)
+	tokens := UsageTokens{InputTokens: 1_000_000, OutputTokens: 1_000_000, CacheReadTokens: 1_000_000}
+
+	for _, model := range []string{
+		"gemini-3.6-flash",
+		"gemini-3.6-flash-high",
+		"gemini-3.6-flash-low",
+		"gemini-3.6-flash-medium",
+		"gemini-3.6-flash-tiered",
+	} {
+		t.Run(model, func(t *testing.T) {
+			cost, err := svc.CalculateCost(model, tokens, 1)
+			require.NoError(t, err)
+			require.InDelta(t, 1.5, cost.InputCost, 1e-12)
+			require.InDelta(t, 7.5, cost.OutputCost, 1e-12)
+			require.InDelta(t, 0.15, cost.CacheReadCost, 1e-12)
+			require.InDelta(t, 9.15, cost.TotalCost, 1e-12)
+		})
+	}
+}
+
+func TestDefaultPricingIncludesGemini36FlashRates(t *testing.T) {
+	data, err := os.ReadFile(filepath.Join("..", "..", "resources", "model-pricing", "model_prices_and_context_window.json"))
+	require.NoError(t, err)
+
+	pricingSvc := &PricingService{}
+	pricingData, err := pricingSvc.parsePricingData(data)
+	require.NoError(t, err)
+	pricingSvc.pricingData = pricingData
+	billingSvc := NewBillingService(&config.Config{}, pricingSvc)
+
+	for _, model := range []string{"gemini-3.6-flash", "gemini-3.6-flash-low", "gemini-3.6-flash-high"} {
+		t.Run(model, func(t *testing.T) {
+			pricing, err := billingSvc.GetModelPricing(model)
+			require.NoError(t, err)
+			require.InDelta(t, 1.5e-6, pricing.InputPricePerToken, 1e-12)
+			require.InDelta(t, 7.5e-6, pricing.OutputPricePerToken, 1e-12)
+			require.InDelta(t, 0.15e-6, pricing.CacheReadPricePerToken, 1e-12)
+		})
+	}
+}
+
 func TestDefaultPricingIncludesCodexAutoReview(t *testing.T) {
 	data, err := os.ReadFile(filepath.Join("..", "..", "resources", "model-pricing", "model_prices_and_context_window.json"))
 	require.NoError(t, err)

backend/resources/model-pricing/model_prices_and_context_window.json

@@ -2339,6 +2339,23 @@
     "supports_web_search": true,
     "web_search_billing_unit": "per_query"
   },
+  "gemini-3.6-flash": {
+    "cache_read_input_token_cost": 1.5e-07,
+    "cache_read_input_token_cost_priority": 2.7e-07,
+    "input_cost_per_token": 1.5e-06,
+    "input_cost_per_token_priority": 2.7e-06,
+    "litellm_provider": "vertex_ai-language-models",
+    "max_input_tokens": 1048576,
+    "max_output_tokens": 65536,
+    "max_tokens": 65536,
+    "mode": "chat",
+    "output_cost_per_token": 7.5e-06,
+    "output_cost_per_token_priority": 1.35e-05,
+    "source": "https://ai.google.dev/gemini-api/docs/latest-model",
+    "supports_prompt_caching": true,
+    "supports_reasoning": true,
+    "supports_service_tier": true
+  },
   "gemini-embedding-001": {
     "input_cost_per_token": 1.5e-07,
     "litellm_provider": "vertex_ai-embedding-models",