feat(gemini): 补齐 Gemini 4 Argon 计费

- 内置兜底价：gemini-4-argon 为 $2 输入 / $10 输出 / $0.10 缓存读取（每百万
  token，优惠价有效期至 2026-11-30，2026-12-01 起为 $4 / $20 / $0.20）；
  Cyber 变体（gemini-4-argon-cyber-permissive）不参与优惠，为 $4 / $20 / $0.20，
  匹配时先于基础型号判断
- Antigravity 的 -low / -medium / -high 档位名归一到基础价
- 本地价格快照新增两个不带前缀的条目，含 Priority 档价格；所有提示词长度同价，
  不带长上下文阶梯；只支持隐式缓存，不带存储价

backend/internal/service/billing_service.go

@@ -524,6 +524,27 @@ func (s *BillingService) initFallbackPricing() {
 		SupportsCacheBreakdown: false,
 	}
 
+	// Gemini 4 Argon (Google AI pricing: $2 input / $10 output / $0.10 cached
+	// input per MTok, promotional through 2026-11-30; official rates double to
+	// $4/$20/$0.20 from 2026-12-01). One rate covers every prompt length.
+	// Antigravity's -high/-low/-medium aliases are matched below so unavailable
+	// remote pricing never records token-bearing requests at $0.
+	s.fallbackPrices["gemini-4-argon"] = &ModelPricing{
+		InputPricePerToken:     2e-6,
+		OutputPricePerToken:    10e-6,
+		CacheReadPricePerToken: 0.1e-6,
+		SupportsCacheBreakdown: false,
+	}
+
+	// Gemini 4 Argon Cyber Permissive (Vertex AI pricing: $4 input / $20 output /
+	// $0.20 cached input per MTok). The Gemini 4 Argon promotion does not apply.
+	s.fallbackPrices["gemini-4-argon-cyber-permissive"] = &ModelPricing{
+		InputPricePerToken:     4e-6,
+		OutputPricePerToken:    20e-6,
+		CacheReadPricePerToken: 0.2e-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
@@ -1082,6 +1103,13 @@ func (s *BillingService) getFallbackPricing(model string) *ModelPricing {
 	if strings.Contains(modelLower, "gemini-3.8-flash") || strings.Contains(modelLower, "gemini-3-8-flash") {
 		return s.fallbackPrices["gemini-3.8-flash"]
 	}
+	// Cyber 变体不享受 Gemini 4 Argon 的优惠价，必须先于 "gemini-4-argon" 判。
+	if strings.Contains(modelLower, "gemini-4-argon-cyber") {
+		return s.fallbackPrices["gemini-4-argon-cyber-permissive"]
+	}
+	if strings.Contains(modelLower, "gemini-4-argon") {
+		return s.fallbackPrices["gemini-4-argon"]
+	}
 
 	// DeepSeek 系列：官方模型 V4 Pro/Flash（含 vision-exp）按各自价卡；
 	// 其余 deepseek-*（含已停服的 deepseek-chat / deepseek-reasoner 与未知型号）

backend/internal/service/gemini_cache_storage_pricing_test.go

@@ -43,7 +43,10 @@ func TestBuiltinGeminiCacheStoragePrice(t *testing.T) {
 		{"gemini-flash-latest", 1e-6},
 		{"gemini-flash-lite-latest", 1e-6},
 		{"gemini-pro-latest", 4.5e-6},
-		// 无存储价
+		// 无存储价（Gemini 4 Argon 仅支持隐式缓存）
+		{"gemini-4-argon", 0},
+		{"gemini-4-argon-high", 0},
+		{"gemini-4-argon-cyber-permissive", 0},
 		{"gemini-embedding-001", 0},
 		{"gemini-unknown-model", 0},
 		{"claude-sonnet-5-5", 0},

backend/internal/service/pricing_service.go

@@ -1306,7 +1306,7 @@ func (s *PricingService) buildModelLookupCandidates(modelLower string) []string
 	normalized := normalizeModelNameForPricing(modelLower)
 
 	// A tier-specific entry should take precedence when the pricing catalog gains
-	// one later. Antigravity's Gemini Flash thinking tiers share the base rate,
+	// one later. Antigravity's Gemini thinking tiers share the base rate,
 	// so the normalized base remains the fallback after the exact aliases.
 	candidates := rawCandidates
 	if normalizeGeminiThinkingTierAlias(lastSegment(modelLower)) != lastSegment(modelLower) {
@@ -1368,12 +1368,12 @@ func normalizeModelNameForPricing(model string) string {
 	return normalizeGeminiThinkingTierAlias(model)
 }
 
-// normalizeGeminiThinkingTierAlias maps Antigravity's Gemini 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 corresponding base model's price card.
+// normalizeGeminiThinkingTierAlias maps Antigravity's Gemini 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 corresponding base model's price card.
 func normalizeGeminiThinkingTierAlias(model string) string {
-	for _, baseModel := range []string{"gemini-3.6-flash", "gemini-3.7-flash", "gemini-3.8-flash"} {
+	for _, baseModel := range []string{"gemini-3.6-flash", "gemini-3.7-flash", "gemini-3.8-flash", "gemini-4-argon"} {
 		for _, tier := range []string{"-high", "-low", "-medium", "-tiered"} {
 			if model == baseModel+tier {
 				return baseModel

backend/internal/service/pricing_service_test.go

@@ -732,6 +732,93 @@ func TestDefaultPricingIncludesGemini36FlashRates(t *testing.T) {
 	}
 }
 
+func TestPricingService_Gemini4ArgonThinkingTiersUseBasePricing(t *testing.T) {
+	basePricing := &LiteLLMModelPricing{InputCostPerToken: 2e-6, OutputCostPerToken: 10e-6, CacheReadInputTokenCost: 0.1e-6}
+	cyberPricing := &LiteLLMModelPricing{InputCostPerToken: 4e-6, OutputCostPerToken: 20e-6, CacheReadInputTokenCost: 0.2e-6}
+	svc := &PricingService{pricingData: map[string]*LiteLLMModelPricing{
+		"gemini-4-argon":                  basePricing,
+		"gemini-4-argon-cyber-permissive": cyberPricing,
+	}}
+
+	for _, model := range []string{
+		"gemini-4-argon",
+		"gemini-4-argon-high",
+		"gemini-4-argon-low",
+		"gemini-4-argon-medium",
+		"models/gemini-4-argon-high",
+		"projects/p/locations/global/publishers/google/models/gemini-4-argon",
+	} {
+		t.Run(model, func(t *testing.T) {
+			require.Same(t, basePricing, svc.GetModelPricing(model))
+		})
+	}
+	require.Same(t, cyberPricing, svc.GetModelPricing("gemini-4-argon-cyber-permissive"))
+}
+
+func TestBillingService_Gemini4ArgonFallbacksAreBillable(t *testing.T) {
+	svc := NewBillingService(&config.Config{}, nil)
+	tokens := UsageTokens{InputTokens: 1_000_000, OutputTokens: 1_000_000, CacheReadTokens: 1_000_000}
+
+	for _, tc := range []struct {
+		model               string
+		input, output, read float64
+	}{
+		{"gemini-4-argon", 2, 10, 0.1},
+		{"gemini-4-argon-high", 2, 10, 0.1},
+		{"gemini-4-argon-low", 2, 10, 0.1},
+		{"gemini-4-argon-medium", 2, 10, 0.1},
+		{"models/gemini-4-argon", 2, 10, 0.1},
+		{"gemini-4-argon-cyber-permissive", 4, 20, 0.2},
+		{"gemini-4-argon-cyber", 4, 20, 0.2},
+		{"publishers/google/models/gemini-4-argon-cyber-permissive", 4, 20, 0.2},
+	} {
+		t.Run(tc.model, func(t *testing.T) {
+			cost, err := svc.CalculateCost(tc.model, tokens, 1)
+			require.NoError(t, err)
+			require.InDelta(t, tc.input, cost.InputCost, 1e-12)
+			require.InDelta(t, tc.output, cost.OutputCost, 1e-12)
+			require.InDelta(t, tc.read, cost.CacheReadCost, 1e-12)
+			require.InDelta(t, tc.input+tc.output+tc.read, cost.TotalCost, 1e-12)
+		})
+	}
+}
+
+func TestDefaultPricingIncludesGemini4ArgonRates(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
+
+	for _, tc := range []struct {
+		model               string
+		provider            string
+		input, output, read float64
+	}{
+		{"gemini-4-argon", "gemini", 2e-6, 10e-6, 0.1e-6},
+		{"gemini-4-argon-high", "gemini", 2e-6, 10e-6, 0.1e-6},
+		{"gemini-4-argon-cyber-permissive", "vertex_ai-language-models", 4e-6, 20e-6, 0.2e-6},
+	} {
+		t.Run(tc.model, func(t *testing.T) {
+			pricing := pricingSvc.GetModelPricing(tc.model)
+			require.NotNil(t, pricing)
+			require.Equal(t, tc.provider, pricing.LiteLLMProvider)
+			require.InDelta(t, tc.input, pricing.InputCostPerToken, 1e-15)
+			require.InDelta(t, tc.output, pricing.OutputCostPerToken, 1e-15)
+			require.InDelta(t, tc.read, pricing.CacheReadInputTokenCost, 1e-15)
+			// Priority 档为标准价的 1.8 倍；单一价档不分提示词长度，显式缓存不可用故无存储价。
+			require.InDelta(t, tc.input*1.8, pricing.InputCostPerTokenPriority, 1e-15)
+			require.InDelta(t, tc.output*1.8, pricing.OutputCostPerTokenPriority, 1e-15)
+			require.InDelta(t, tc.read*1.8, pricing.CacheReadInputTokenCostPriority, 1e-15)
+			require.Zero(t, pricing.LongContextInputTokenThreshold)
+			require.Zero(t, pricing.CacheStorageCostPerTokenPerHour)
+		})
+	}
+	require.Contains(t, pricingSvc.ListModelNamesByProvider("gemini"), "gemini-4-argon")
+}
+
 func TestDefaultPricingUsesCurrentCodexAutoReviewBaseRates(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

@@ -2666,6 +2666,40 @@
     "supports_reasoning": true,
     "supports_service_tier": true
   },
+  "gemini-4-argon": {
+    "cache_read_input_token_cost": 1e-07,
+    "cache_read_input_token_cost_priority": 1.8e-07,
+    "input_cost_per_token": 2e-06,
+    "input_cost_per_token_priority": 3.6e-06,
+    "litellm_provider": "gemini",
+    "max_input_tokens": 1048576,
+    "max_output_tokens": 65536,
+    "max_tokens": 65536,
+    "mode": "chat",
+    "output_cost_per_token": 1e-05,
+    "output_cost_per_token_priority": 1.8e-05,
+    "source": "https://ai.google.dev/gemini-api/docs/pricing",
+    "supports_prompt_caching": true,
+    "supports_reasoning": true,
+    "supports_service_tier": true
+  },
+  "gemini-4-argon-cyber-permissive": {
+    "cache_read_input_token_cost": 2e-07,
+    "cache_read_input_token_cost_priority": 3.6e-07,
+    "input_cost_per_token": 4e-06,
+    "input_cost_per_token_priority": 7.2e-06,
+    "litellm_provider": "vertex_ai-language-models",
+    "max_input_tokens": 1048576,
+    "max_output_tokens": 65536,
+    "max_tokens": 65536,
+    "mode": "chat",
+    "output_cost_per_token": 2e-05,
+    "output_cost_per_token_priority": 3.6e-05,
+    "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
+    "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",