feat: support cached input token billing

apps/api/internal/clients/clients_test.go

@@ -631,7 +631,7 @@ func TestOpenAIClientChatStreamContract(t *testing.T) {
 		w.Header().Set("Content-Type", "text/event-stream")
 		_, _ = w.Write([]byte("data: {\"id\":\"chatcmpl-stream\",\"object\":\"chat.completion.chunk\",\"model\":\"deepseek-v4-flash\",\"choices\":[{\"delta\":{\"content\":\"hello\"}}],\"usage\":null}\n\n"))
 		_, _ = w.Write([]byte("data: {\"id\":\"chatcmpl-stream\",\"object\":\"chat.completion.chunk\",\"model\":\"deepseek-v4-flash\",\"choices\":[{\"delta\":{\"content\":\" world\"},\"finish_reason\":\"stop\"}],\"usage\":null}\n\n"))
-		_, _ = w.Write([]byte("data: {\"id\":\"chatcmpl-stream\",\"object\":\"chat.completion.chunk\",\"model\":\"deepseek-v4-flash\",\"choices\":[],\"usage\":{\"prompt_tokens\":1,\"completion_tokens\":2,\"total_tokens\":3}}\n\n"))
+		_, _ = w.Write([]byte("data: {\"id\":\"chatcmpl-stream\",\"object\":\"chat.completion.chunk\",\"model\":\"deepseek-v4-flash\",\"choices\":[],\"usage\":{\"prompt_tokens\":1,\"completion_tokens\":2,\"total_tokens\":3,\"prompt_tokens_details\":{\"cached_tokens\":1}}}\n\n"))
 		_, _ = w.Write([]byte("data: [DONE]\n\n"))
 	}))
 	defer server.Close()
@@ -662,6 +662,9 @@ func TestOpenAIClientChatStreamContract(t *testing.T) {
 	if response.Usage.TotalTokens != 3 {
 		t.Fatalf("unexpected usage: %+v", response.Usage)
 	}
+	if response.Usage.CachedInputTokens != 1 {
+		t.Fatalf("expected cached input tokens from stream usage, got %+v", response.Usage)
+	}
 	choices, _ := response.Result["choices"].([]any)
 	choice, _ := choices[0].(map[string]any)
 	message, _ := choice["message"].(map[string]any)
@@ -670,6 +673,111 @@ func TestOpenAIClientChatStreamContract(t *testing.T) {
 	}
 }
 
+func TestOpenAIClientChatStreamExtractsCachedInputTokensFromProviderUsageShapes(t *testing.T) {
+	cases := []struct {
+		name   string
+		usage  map[string]any
+		cached int
+	}{
+		{
+			name: "aliyun dashscope openai-compatible",
+			usage: map[string]any{
+				"prompt_tokens":     10,
+				"completion_tokens": 2,
+				"total_tokens":      12,
+				"prompt_tokens_details": map[string]any{
+					"cached_tokens": 7,
+				},
+			},
+			cached: 7,
+		},
+		{
+			name: "volces responses-style input details",
+			usage: map[string]any{
+				"prompt_tokens":     9,
+				"completion_tokens": 3,
+				"total_tokens":      12,
+				"input_tokens_details": map[string]any{
+					"cached_tokens": 6,
+				},
+			},
+			cached: 6,
+		},
+		{
+			name: "deepseek prompt cache hit tokens",
+			usage: map[string]any{
+				"prompt_tokens":            8,
+				"completion_tokens":        4,
+				"total_tokens":             12,
+				"prompt_cache_hit_tokens":  5,
+				"prompt_cache_miss_tokens": 3,
+			},
+			cached: 5,
+		},
+		{
+			name: "glm openai-compatible",
+			usage: map[string]any{
+				"prompt_tokens":     11,
+				"completion_tokens": 1,
+				"total_tokens":      12,
+				"prompt_tokens_details": map[string]any{
+					"cached_tokens": 4,
+				},
+			},
+			cached: 4,
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			var gotIncludeUsage bool
+			server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				var body map[string]any
+				if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+					t.Fatalf("decode request: %v", err)
+				}
+				streamOptions, _ := body["stream_options"].(map[string]any)
+				gotIncludeUsage, _ = streamOptions["include_usage"].(bool)
+				usageBytes, _ := json.Marshal(tc.usage)
+				w.Header().Set("Content-Type", "text/event-stream")
+				_, _ = w.Write([]byte("data: {\"id\":\"chatcmpl-cache\",\"object\":\"chat.completion.chunk\",\"model\":\"cache-test\",\"choices\":[{\"delta\":{\"content\":\"ok\"},\"finish_reason\":\"stop\"}],\"usage\":null}\n\n"))
+				_, _ = w.Write([]byte("data: {\"id\":\"chatcmpl-cache\",\"object\":\"chat.completion.chunk\",\"model\":\"cache-test\",\"choices\":[],\"usage\":" + string(usageBytes) + "}\n\n"))
+				_, _ = w.Write([]byte("data: [DONE]\n\n"))
+			}))
+			defer server.Close()
+
+			response, err := (OpenAIClient{HTTPClient: server.Client()}).Run(context.Background(), Request{
+				Kind:  "chat.completions",
+				Model: "Cache-Test",
+				Body: map[string]any{
+					"model":    "Cache-Test",
+					"messages": []any{map[string]any{"role": "user", "content": "ping"}},
+					"stream":   true,
+				},
+				Candidate: store.RuntimeModelCandidate{
+					BaseURL:     server.URL,
+					ModelName:   "cache-test",
+					Credentials: map[string]any{"apiKey": "test-key"},
+				},
+			})
+			if err != nil {
+				t.Fatalf("run openai stream client: %v", err)
+			}
+			if !gotIncludeUsage {
+				t.Fatal("expected upstream stream_options.include_usage=true")
+			}
+			if response.Usage.CachedInputTokens != tc.cached {
+				t.Fatalf("expected cached input tokens %d, got %+v", tc.cached, response.Usage)
+			}
+			resultUsage, _ := response.Result["usage"].(map[string]any)
+			promptDetails, _ := resultUsage["prompt_tokens_details"].(map[string]any)
+			if intFromAny(promptDetails["cached_tokens"]) != tc.cached {
+				t.Fatalf("result usage should expose normalized cached_tokens, got %+v", resultUsage)
+			}
+		})
+	}
+}
+
 func TestOpenAIClientChatStreamPreservesStructuredDeltas(t *testing.T) {
 	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		w.Header().Set("Content-Type", "text/event-stream")

apps/api/internal/clients/gemini.go

@@ -519,7 +519,11 @@ func geminiUsage(raw map[string]any) Usage {
 	input := intFromAny(usageMap["prompt_tokens"])
 	output := intFromAny(usageMap["completion_tokens"])
 	total := intFromAny(usageMap["total_tokens"])
-	return Usage{InputTokens: input, OutputTokens: output, TotalTokens: total}
+	cachedInput := cachedInputTokensFromOpenAIUsage(usageMap)
+	if cachedInput > input && input > 0 {
+		cachedInput = input
+	}
+	return Usage{InputTokens: input, OutputTokens: output, CachedInputTokens: cachedInput, TotalTokens: total}
 }
 
 func geminiUsageMap(raw map[string]any) map[string]any {
@@ -527,8 +531,13 @@ func geminiUsageMap(raw map[string]any) map[string]any {
 	input := intFromAny(meta["promptTokenCount"])
 	output := intFromAny(meta["candidatesTokenCount"])
 	total := intFromAny(meta["totalTokenCount"])
+	cachedInput := intFromAny(firstPresent(meta["cachedContentTokenCount"], meta["cached_content_token_count"]))
 	if total == 0 {
 		total = input + output
 	}
-	return map[string]any{"prompt_tokens": input, "completion_tokens": output, "total_tokens": total}
+	usage := map[string]any{"prompt_tokens": input, "completion_tokens": output, "total_tokens": total}
+	if cachedInput > 0 {
+		usage["prompt_tokens_details"] = map[string]any{"cached_tokens": cachedInput}
+	}
+	return usage
 }

apps/api/internal/clients/helpers.go

@@ -185,11 +185,17 @@ func buildOpenAIStreamResult(last map[string]any, parts []string, reasoningParts
 		}},
 	}
 	if usage.TotalTokens > 0 {
-		out["usage"] = map[string]any{
+		usageMap := map[string]any{
 			"prompt_tokens":     usage.InputTokens,
 			"completion_tokens": usage.OutputTokens,
 			"total_tokens":      usage.TotalTokens,
 		}
+		if usage.CachedInputTokens > 0 {
+			usageMap["prompt_tokens_details"] = map[string]any{
+				"cached_tokens": usage.CachedInputTokens,
+			}
+		}
+		out["usage"] = usageMap
 	}
 	return out
 }
@@ -877,7 +883,42 @@ func usageFromOpenAI(result map[string]any) Usage {
 	if total == 0 {
 		total = input + output
 	}
-	return Usage{InputTokens: input, OutputTokens: output, TotalTokens: total}
+	cachedInput := cachedInputTokensFromOpenAIUsage(usage)
+	if cachedInput > input && input > 0 {
+		cachedInput = input
+	}
+	return Usage{InputTokens: input, OutputTokens: output, CachedInputTokens: cachedInput, TotalTokens: total}
+}
+
+func cachedInputTokensFromOpenAIUsage(usage map[string]any) int {
+	if len(usage) == 0 {
+		return 0
+	}
+	promptDetails, _ := firstPresent(usage["prompt_tokens_details"], usage["promptTokensDetails"]).(map[string]any)
+	inputDetails, _ := firstPresent(usage["input_tokens_details"], usage["inputTokensDetails"]).(map[string]any)
+	usageMetadata, _ := usage["usageMetadata"].(map[string]any)
+	return intFromAny(firstPresent(
+		promptDetails["cached_tokens"],
+		promptDetails["cachedTokens"],
+		promptDetails["cache_read_input_tokens"],
+		promptDetails["cacheReadInputTokens"],
+		inputDetails["cached_tokens"],
+		inputDetails["cachedTokens"],
+		inputDetails["cache_read_input_tokens"],
+		inputDetails["cacheReadInputTokens"],
+		usage["prompt_cache_hit_tokens"],
+		usage["promptCacheHitTokens"],
+		usage["cached_tokens"],
+		usage["cachedTokens"],
+		usage["cached_input_tokens"],
+		usage["cachedInputTokens"],
+		usage["cache_read_input_tokens"],
+		usage["cacheReadInputTokens"],
+		usage["cached_content_token_count"],
+		usage["cachedContentTokenCount"],
+		usageMetadata["cached_content_token_count"],
+		usageMetadata["cachedContentTokenCount"],
+	))
 }
 
 func requestIDFromHTTPResponse(resp *http.Response) string {

apps/api/internal/clients/types.go

@@ -36,9 +36,10 @@ type Response struct {
 }
 
 type Usage struct {
-	InputTokens  int
+	InputTokens       int
-	OutputTokens int
+	OutputTokens      int
-	TotalTokens  int
+	CachedInputTokens int
+	TotalTokens       int
 }
 
 type Progress struct {

apps/api/internal/httpapi/chat_completions_mode_test.go

@@ -165,6 +165,87 @@ func TestWriteCompatibleTaskResponseReturnsSSEWhenStreamIsTrue(t *testing.T) {
 	}
 }
 
+func TestWriteCompatibleTaskResponseStreamsCachedUsageProviderShapes(t *testing.T) {
+	cases := []struct {
+		name     string
+		usage    map[string]any
+		fragment string
+	}{
+		{
+			name: "aliyun dashscope",
+			usage: map[string]any{
+				"prompt_tokens":     10,
+				"completion_tokens": 2,
+				"total_tokens":      12,
+				"prompt_tokens_details": map[string]any{
+					"cached_tokens": 7,
+				},
+			},
+			fragment: `"prompt_tokens_details":{"cached_tokens":7}`,
+		},
+		{
+			name: "volces responses",
+			usage: map[string]any{
+				"prompt_tokens":     9,
+				"completion_tokens": 3,
+				"total_tokens":      12,
+				"input_tokens_details": map[string]any{
+					"cached_tokens": 6,
+				},
+			},
+			fragment: `"input_tokens_details":{"cached_tokens":6}`,
+		},
+		{
+			name: "deepseek",
+			usage: map[string]any{
+				"prompt_tokens":            8,
+				"completion_tokens":        4,
+				"total_tokens":             12,
+				"prompt_cache_hit_tokens":  5,
+				"prompt_cache_miss_tokens": 3,
+			},
+			fragment: `"prompt_cache_hit_tokens":5`,
+		},
+		{
+			name: "glm",
+			usage: map[string]any{
+				"prompt_tokens":     11,
+				"completion_tokens": 1,
+				"total_tokens":      12,
+				"prompt_tokens_details": map[string]any{
+					"cached_tokens": 4,
+				},
+			},
+			fragment: `"prompt_tokens_details":{"cached_tokens":4}`,
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			executor := &fakeTaskExecutor{
+				deltas: []clients.StreamDeltaEvent{{Text: "ok"}},
+				output: map[string]any{
+					"id":     "chatcmpl-cache",
+					"object": "chat.completion",
+					"usage":  tc.usage,
+				},
+			}
+			req := httptest.NewRequest(http.MethodPost, "/api/v1/chat/completions", nil)
+			recorder := httptest.NewRecorder()
+
+			writeCompatibleTaskResponse(context.Background(), recorder, req, executor, "chat.completions", "cache-test", store.GatewayTask{ID: "task-test"}, &auth.User{}, true, true)
+
+			body := recorder.Body.String()
+			if !strings.Contains(body, tc.fragment) {
+				t.Fatalf("SSE body missing cached usage fragment %s: %s", tc.fragment, body)
+			}
+			if !strings.Contains(body, `"choices":[]`) || !strings.Contains(body, "data: [DONE]") {
+				t.Fatalf("SSE body should include final usage chunk and done marker: %s", body)
+			}
+		})
+	}
+}
+
 func TestWriteCompatibleTaskResponseStreamsStructuredToolAndReasoningDeltas(t *testing.T) {
 	executor := &fakeTaskExecutor{
 		deltas: []clients.StreamDeltaEvent{

apps/api/internal/httpapi/model_catalog.go

@@ -759,7 +759,7 @@ func billingConfigLines(config map[string]any) []string {
 
 func textPricingResource(config map[string]any) bool {
 	switch strings.ToLower(stringValue(config["resourceType"])) {
-	case "text", "text_total", "text_input", "text_output":
+	case "text", "text_total", "text_input", "text_cached_input", "text_output":
 		return true
 	default:
 		return false
@@ -771,6 +771,7 @@ func textPricingLines(config map[string]any) []string {
 		return nil
 	}
 	inputPrice, hasInput := textPriceFromConfig(config, []string{"textInputPer1k", "textInput", "text_input", "inputTokenPrice", "inputPrice"})
+	cachedInputPrice, hasCachedInput := textPriceFromConfig(config, []string{"textCachedInputPer1k", "textCachedInput", "text_cached_input", "cachedInputTokenPrice", "inputCacheHitTokenPrice", "cachedInputPrice"})
 	outputPrice, hasOutput := textPriceFromConfig(config, []string{"textOutputPer1k", "textOutput", "text_output", "outputTokenPrice", "outputPrice"})
 
 	for _, key := range []string{"formulaConfig", "formula_config"} {
@@ -778,6 +779,9 @@ func textPricingLines(config map[string]any) []string {
 		if !hasInput {
 			inputPrice, hasInput = textPriceFromConfig(formulaConfig, []string{"inputTokenPrice", "textInputPer1k", "textInput", "text_input", "inputPrice"})
 		}
+		if !hasCachedInput {
+			cachedInputPrice, hasCachedInput = textPriceFromConfig(formulaConfig, []string{"cachedInputTokenPrice", "textCachedInputPer1k", "textCachedInput", "text_cached_input", "inputCacheHitTokenPrice", "cachedInputPrice"})
+		}
 		if !hasOutput {
 			outputPrice, hasOutput = textPriceFromConfig(formulaConfig, []string{"outputTokenPrice", "textOutputPer1k", "textOutput", "text_output", "outputPrice"})
 		}
@@ -791,6 +795,9 @@ func textPricingLines(config map[string]any) []string {
 		if !hasInput {
 			inputPrice, hasInput = textPriceFromConfig(nested, []string{"inputTokenPrice", "textInputPer1k", "textInput", "text_input", "inputPrice"})
 		}
+		if !hasCachedInput {
+			cachedInputPrice, hasCachedInput = textPriceFromConfig(nested, []string{"cachedInputTokenPrice", "textCachedInputPer1k", "textCachedInput", "text_cached_input", "inputCacheHitTokenPrice", "cachedInputPrice"})
+		}
 		if !hasOutput {
 			outputPrice, hasOutput = textPriceFromConfig(nested, []string{"outputTokenPrice", "textOutputPer1k", "textOutput", "text_output", "outputPrice"})
 		}
@@ -799,6 +806,9 @@ func textPricingLines(config map[string]any) []string {
 			if !hasInput {
 				inputPrice, hasInput = textPriceFromConfig(formulaConfig, []string{"inputTokenPrice", "textInputPer1k", "textInput", "text_input", "inputPrice"})
 			}
+			if !hasCachedInput {
+				cachedInputPrice, hasCachedInput = textPriceFromConfig(formulaConfig, []string{"cachedInputTokenPrice", "textCachedInputPer1k", "textCachedInput", "text_cached_input", "inputCacheHitTokenPrice", "cachedInputPrice"})
+			}
 			if !hasOutput {
 				outputPrice, hasOutput = textPriceFromConfig(formulaConfig, []string{"outputTokenPrice", "textOutputPer1k", "textOutput", "text_output", "outputPrice"})
 			}
@@ -824,6 +834,10 @@ func textPricingLines(config map[string]any) []string {
 			if !hasInput {
 				inputPrice, hasInput = basePrice, true
 			}
+		case "text_cached_input":
+			if !hasCachedInput {
+				cachedInputPrice, hasCachedInput = basePrice, true
+			}
 		case "text_output":
 			if !hasOutput {
 				outputPrice, hasOutput = basePrice, true
@@ -835,6 +849,9 @@ func textPricingLines(config map[string]any) []string {
 	if hasInput {
 		lines = append(lines, "输入 "+formatNumber(inputPrice)+"/k tokens")
 	}
+	if hasCachedInput {
+		lines = append(lines, "缓存输入 "+formatNumber(cachedInputPrice)+"/k tokens")
+	}
 	if hasOutput {
 		lines = append(lines, "输出 "+formatNumber(outputPrice)+"/k tokens")
 	}

apps/api/internal/httpapi/openapi_models.go

@@ -294,9 +294,19 @@ type ChatCompletionChoiceMessage struct {
 }
 
 type ChatCompletionUsage struct {
-	PromptTokens     int `json:"prompt_tokens,omitempty" example:"12"`
+	PromptTokens        int                      `json:"prompt_tokens,omitempty" example:"12"`
-	CompletionTokens int `json:"completion_tokens,omitempty" example:"8"`
+	CompletionTokens    int                      `json:"completion_tokens,omitempty" example:"8"`
-	TotalTokens      int `json:"total_tokens,omitempty" example:"20"`
+	TotalTokens         int                      `json:"total_tokens,omitempty" example:"20"`
+	PromptTokensDetails *ChatPromptTokensDetails `json:"prompt_tokens_details,omitempty"`
+	InputTokensDetails  *ChatInputTokensDetails  `json:"input_tokens_details,omitempty"`
+}
+
+type ChatPromptTokensDetails struct {
+	CachedTokens int `json:"cached_tokens,omitempty" example:"4"`
+}
+
+type ChatInputTokensDetails struct {
+	CachedTokens int `json:"cached_tokens,omitempty" example:"4"`
 }
 
 type NetworkProxyConfigResponse struct {

apps/api/internal/runner/limits_test.go

@@ -1,6 +1,7 @@
 package runner
 
 import (
+	"context"
 	"testing"
 
 	"github.com/easyai/easyai-ai-gateway/apps/api/internal/clients"
@@ -29,6 +30,44 @@ func TestTokenUsageAmountsFallsBackToInputOutputTotal(t *testing.T) {
 	}
 }
 
+func TestBillingsSplitsCachedInputTokens(t *testing.T) {
+	service := &Service{}
+	candidate := store.RuntimeModelCandidate{
+		ModelName: "test-model",
+		BillingConfig: map[string]any{
+			"textInputPer1k":       1.0,
+			"textCachedInputPer1k": 0.25,
+			"textOutputPer1k":      2.0,
+		},
+	}
+	response := clients.Response{
+		Usage: clients.Usage{
+			InputTokens:       1000,
+			CachedInputTokens: 400,
+			OutputTokens:      500,
+			TotalTokens:       1500,
+		},
+	}
+
+	lines := service.billings(context.Background(), nil, "chat.completions", nil, candidate, response, false)
+
+	if len(lines) != 3 {
+		t.Fatalf("expected uncached input, cached input, and output lines, got %+v", lines)
+	}
+	uncached, _ := lines[0].(map[string]any)
+	cached, _ := lines[1].(map[string]any)
+	output, _ := lines[2].(map[string]any)
+	if uncached["resourceType"] != "text_input" || uncached["quantity"] != 600 || uncached["amount"] != 0.6 {
+		t.Fatalf("unexpected uncached input billing line: %+v", uncached)
+	}
+	if cached["resourceType"] != "text_cached_input" || cached["quantity"] != 400 || cached["amount"] != 0.1 {
+		t.Fatalf("unexpected cached input billing line: %+v", cached)
+	}
+	if output["resourceType"] != "text_output" || output["quantity"] != 500 || output["amount"] != 1.0 {
+		t.Fatalf("unexpected output billing line: %+v", output)
+	}
+}
+
 func TestEffectiveRateLimitPolicyTreatsEmptyRuntimePolicyAsUnlimited(t *testing.T) {
 	policy := effectiveRateLimitPolicy(store.RuntimeModelCandidate{
 		PlatformRateLimitPolicy: map[string]any{"rules": []any{

apps/api/internal/runner/pricing.go

@@ -62,6 +62,7 @@ func (s *Service) billings(ctx context.Context, user *auth.User, kind string, bo
 	if isTextBillingKind(kind) {
 		inputTokens := response.Usage.InputTokens
 		outputTokens := response.Usage.OutputTokens
+		cachedInputTokens := response.Usage.CachedInputTokens
 		if isTextInputOnlyKind(kind) && inputTokens == 0 && response.Usage.TotalTokens > 0 {
 			inputTokens = response.Usage.TotalTokens
 		}
@@ -71,8 +72,37 @@ func (s *Service) billings(ctx context.Context, user *auth.User, kind string, bo
 				outputTokens = 1
 			}
 		}
-		inputAmount := roundPrice(float64(inputTokens) / 1000 * resourcePrice(config, "text", "textInputPer1k", "inputTokenPrice", "basePrice") * discount)
+		if cachedInputTokens > inputTokens && inputTokens > 0 {
-		lines := []any{billingLine(candidate, "text_input", "1k_tokens", inputTokens, inputAmount, discount, simulated)}
+			cachedInputTokens = inputTokens
+		}
+		uncachedInputTokens := inputTokens - cachedInputTokens
+		if uncachedInputTokens < 0 {
+			uncachedInputTokens = 0
+		}
+		inputPrice := resourcePrice(config, "text", "textInputPer1k", "inputTokenPrice", "basePrice")
+		cachedInputPrice := resourcePrice(config, "text", "textCachedInputPer1k", "cachedInputTokenPrice", "cacheInputTokenPrice", "textInputCacheHitPer1k", "inputCacheHitTokenPrice")
+		if cachedInputPrice <= 0 {
+			cachedInputPrice = inputPrice
+		}
+		inputAmount := roundPrice(float64(uncachedInputTokens) / 1000 * inputPrice * discount)
+		lines := []any{}
+		if uncachedInputTokens > 0 || cachedInputTokens == 0 {
+			lines = append(lines, billingLineWithDetails(candidate, "text_input", "1k_tokens", uncachedInputTokens, inputAmount, discount, simulated, map[string]any{
+				"inputTokens":         inputTokens,
+				"uncachedInputTokens": uncachedInputTokens,
+				"cachedInputTokens":   cachedInputTokens,
+				"pricePer1k":          inputPrice,
+			}))
+		}
+		if cachedInputTokens > 0 {
+			cachedInputAmount := roundPrice(float64(cachedInputTokens) / 1000 * cachedInputPrice * discount)
+			lines = append(lines, billingLineWithDetails(candidate, "text_cached_input", "1k_tokens", cachedInputTokens, cachedInputAmount, discount, simulated, map[string]any{
+				"inputTokens":         inputTokens,
+				"uncachedInputTokens": uncachedInputTokens,
+				"cachedInputTokens":   cachedInputTokens,
+				"pricePer1k":          cachedInputPrice,
+			}))
+		}
 		if isTextGenerationKind(kind) {
 			outputAmount := roundPrice(float64(outputTokens) / 1000 * resourcePrice(config, "text", "textOutputPer1k", "outputTokenPrice", "basePrice") * discount)
 			lines = append(lines, billingLine(candidate, "text_output", "1k_tokens", outputTokens, outputAmount, discount, simulated))
@@ -234,10 +264,22 @@ func resourcePrice(config map[string]any, resource string, keys ...string) float
 				return value
 			}
 		}
+		if formulaConfig, ok := resourceConfig["formulaConfig"].(map[string]any); ok {
+			for _, key := range keys {
+				if value := floatFromAny(formulaConfig[key]); value > 0 {
+					return value
+				}
+			}
+		}
 		if value := floatFromAny(resourceConfig["basePrice"]); value > 0 {
 			return value
 		}
 	}
+	if resource == "text" {
+		if value := resourcePrice(config, "text_total", keys...); value > 0 {
+			return value
+		}
+	}
 	if resource == "image_edit" {
 		return resourcePrice(config, "image", keys...)
 	}

apps/api/internal/runner/recording.go

@@ -151,6 +151,10 @@ func usageToMap(usage clients.Usage) map[string]any {
 		out["inputTokens"] = usage.InputTokens
 		out["promptTokens"] = usage.InputTokens
 	}
+	if usage.CachedInputTokens > 0 {
+		out["cachedInputTokens"] = usage.CachedInputTokens
+		out["cachedPromptTokens"] = usage.CachedInputTokens
+	}
 	if usage.OutputTokens > 0 {
 		out["outputTokens"] = usage.OutputTokens
 		out["completionTokens"] = usage.OutputTokens

apps/api/internal/store/model_billing_filter.go

@@ -94,7 +94,7 @@ func inferBillingResources(modelType string, resources map[string]bool) {
 
 func billingConfigKeyAllowed(key string, resources map[string]bool) bool {
 	switch normalizeBillingConfigKey(key) {
-	case "text", "texttotal", "text_total", "textinputper1k", "textoutputper1k", "textinput", "textoutput", "text_input", "text_output", "inputtokenprice", "outputtokenprice":
+	case "text", "texttotal", "text_total", "textinputper1k", "textcachedinputper1k", "textoutputper1k", "textinput", "textcachedinput", "textoutput", "text_input", "text_cached_input", "text_output", "inputtokenprice", "cachedinputtokenprice", "inputcachehittokenprice", "outputtokenprice":
 		return resources["text"]
 	case "image", "imagebase":
 		return resources["image"]

apps/api/internal/store/model_billing_filter_test.go

@@ -60,23 +60,25 @@ func TestFilterPlatformModelBillingConfigKeepsTextFlatPricing(t *testing.T) {
 	model := PlatformModel{
 		ModelType: StringList{"text_generate"},
 		BillingConfig: map[string]any{
-			"textInputPer1k":  0.01,
+			"textInputPer1k":       0.01,
-			"textOutputPer1k": 0.02,
+			"textCachedInputPer1k": 0.003,
+			"textOutputPer1k":      0.02,
 			"text_total": map[string]any{
 				"basePrice":       0.01,
-				"formulaConfig":   map[string]any{"inputTokenPrice": 0.01, "outputTokenPrice": 0.02},
+				"formulaConfig":   map[string]any{"inputTokenPrice": 0.01, "cachedInputTokenPrice": 0.003, "outputTokenPrice": 0.02},
 				"dynamicWeight":   map[string]any{"cached": 0.5},
 				"dimensionSchema": map[string]any{"unit": "1k_tokens"},
 			},
-			"inputTokenPrice":  0.01,
+			"inputTokenPrice":       0.01,
-			"outputTokenPrice": 0.02,
+			"cachedInputTokenPrice": 0.003,
-			"image":            map[string]any{"basePrice": 10},
+			"outputTokenPrice":      0.02,
+			"image":                 map[string]any{"basePrice": 10},
 		},
 	}
 
 	filtered := FilterPlatformModelBillingConfig(model)
 
-	assertHasKeys(t, filtered.BillingConfig, "textInputPer1k", "textOutputPer1k", "text_total", "inputTokenPrice", "outputTokenPrice")
+	assertHasKeys(t, filtered.BillingConfig, "textInputPer1k", "textCachedInputPer1k", "textOutputPer1k", "text_total", "inputTokenPrice", "cachedInputTokenPrice", "outputTokenPrice")
 	assertMissingKeys(t, filtered.BillingConfig, "image")
 }
 

apps/api/internal/store/pricing_rules.go

@@ -207,6 +207,8 @@ ORDER BY priority ASC, resource_type ASC`, id)
 		switch resourceType {
 		case "text_input":
 			config["textInputPer1k"] = basePrice
+		case "text_cached_input":
+			config["textCachedInputPer1k"] = basePrice
 		case "text_output":
 			config["textOutputPer1k"] = basePrice
 		case "text_total":
@@ -215,6 +217,9 @@ ORDER BY priority ASC, resource_type ASC`, id)
 				inputPrice = value
 			}
 			config["textInputPer1k"] = inputPrice
+			if cachedInputPrice, ok := pricingRuleNumberFromKeys(formulaConfig, "cachedInputTokenPrice", "cached_input_token_price", "textCachedInputPer1k", "text_cached_input", "inputCacheHitTokenPrice", "input_cache_hit_token_price"); ok {
+				config["textCachedInputPer1k"] = cachedInputPrice
+			}
 			if outputPrice, ok := pricingRuleNumberFromKeys(formulaConfig, "outputTokenPrice", "output_token_price", "textOutputPer1k", "text_output"); ok {
 				config["textOutputPer1k"] = outputPrice
 			}

apps/web/src/pages/WorkspacePage.tsx

@@ -1561,18 +1561,57 @@ function transactionChargeCurrency(transaction: GatewayWalletTransaction) {
 }
 
 function formatTokenUsage(usage: Record<string, unknown>) {
-  const input = tokenValue(usage.inputTokens ?? usage.promptTokens ?? usage.input_tokens ?? usage.prompt_tokens);
+  const input = tokenValue(
-  const output = tokenValue(usage.outputTokens ?? usage.completionTokens ?? usage.output_tokens ?? usage.completion_tokens);
+    usage.inputTokens ??
-  const total = tokenValue(usage.totalTokens ?? usage.total_tokens ?? (input !== null && output !== null ? input + output : null));
+      usage.promptTokens ??
-  if (input === null && output === null && total === null) return '-';
+      usage.input_tokens ??
+      usage.prompt_tokens,
+  );
+  const promptDetails = usageObject(
+    usage.prompt_tokens_details ?? usage.promptTokensDetails,
+  );
+  const inputDetails = usageObject(
+    usage.input_tokens_details ?? usage.inputTokensDetails,
+  );
+  const cachedInput = tokenValue(
+    usage.cachedInputTokens ??
+      usage.cachedPromptTokens ??
+      usage.cached_input_tokens ??
+      usage.cached_tokens ??
+      promptDetails.cached_tokens ??
+      promptDetails.cachedTokens ??
+      inputDetails.cached_tokens ??
+      inputDetails.cachedTokens,
+  );
+  const output = tokenValue(
+    usage.outputTokens ??
+      usage.completionTokens ??
+      usage.output_tokens ??
+      usage.completion_tokens,
+  );
+  const total = tokenValue(
+    usage.totalTokens ??
+      usage.total_tokens ??
+      (input !== null && output !== null ? input + output : null),
+  );
+  if (input === null && cachedInput === null && output === null && total === null) {
+    return '-';
+  }
   return (
     <span className="taskRecordTokenUsage">
       <span>输入：{formatCellValue(input)}/输出：{formatCellValue(output)}</span>
+      {cachedInput !== null ? <span>缓存输入：{formatCellValue(cachedInput)}</span> : null}
       <span>总计：{formatCellValue(total)}</span>
     </span>
   );
 }
 
+function usageObject(value: unknown): Record<string, unknown> {
+  return value && typeof value === 'object' && !Array.isArray(value)
+    ? (value as Record<string, unknown>)
+    : {};
+}
+
 function tokenValue(value: unknown) {
   if (value === undefined || value === null || value === '') return null;
   const numericValue = typeof value === 'number' ? value : Number(value);

apps/web/src/pages/admin/PricingRuleVisualEditor.tsx

@@ -30,9 +30,10 @@ const modeDefinitions: ModeDefinition[] = [
   {
     key: 'text',
     label: '文本',
-    formula: '一次对话扣费 = 输入 Token / 1000 × 输入扣费单价 + 输出 Token / 1000 × 输出扣费单价。',
+    formula:
-    match: (rule) => rule.resourceType.startsWith('text_'),
+      '一次对话扣费 = 未命中缓存输入 Token / 1000 × 输入扣费单价 + 命中缓存输入 Token / 1000 × 缓存输入扣费单价 + 输出 Token / 1000 × 输出扣费单价。',
-    templates: (currency) => [createTextRule(currency, 0.01, 0.03)],
+    match: isTextPricingRule,
+    templates: (currency) => [createTextRule(currency, 0.01, 0.01, 0.03)],
     parameterGroups: [],
   },
   {
@@ -160,7 +161,7 @@ function summarizePricingMode(mode: ModeDefinition, rules: PricingRuleInput[]) {
   const rule = rules[0];
   if (mode.key === 'text') {
     const prices = textPrices(rule);
-    return `文本按输入/输出 Token 分别计费，输入 ${prices.inputTokenPrice}/${unitLabel(rule.unit)}，输出 ${prices.outputTokenPrice}/${unitLabel(rule.unit)}。`;
+    return `文本按输入/缓存输入/输出 Token 分别计费，输入 ${prices.inputTokenPrice}/${unitLabel(rule.unit)}，缓存输入 ${prices.cachedInputTokenPrice}/${unitLabel(rule.unit)}，输出 ${prices.outputTokenPrice}/${unitLabel(rule.unit)}。`;
   }
   const groupSummaries = mode.parameterGroups
     .map((group) => summarizeWeightGroup(group, readGroup(rule.dynamicWeight, group)))
@@ -260,7 +261,11 @@ function TextRuleEditor(props: {
 }) {
   const prices = textPrices(props.rule);
 
-  function updatePrices(inputTokenPrice: number, outputTokenPrice: number) {
+  function updatePrices(
+    inputTokenPrice: number,
+    cachedInputTokenPrice: number,
+    outputTokenPrice: number,
+  ) {
     props.onChange({
       ...props.rule,
       basePrice: inputTokenPrice,
@@ -272,11 +277,17 @@ function TextRuleEditor(props: {
         ...(props.rule.formulaConfig ?? {}),
         formula: textFormula,
         inputTokenPrice,
+        cachedInputTokenPrice,
         outputTokenPrice,
       },
       dimensionSchema: {
         ...(props.rule.dimensionSchema ?? {}),
-        metrics: ['input_tokens', 'output_tokens'],
+        metrics: [
+          'input_tokens',
+          'uncached_input_tokens',
+          'cached_input_tokens',
+          'output_tokens',
+        ],
         unitScale: 1000,
       },
     });
@@ -301,10 +312,52 @@ function TextRuleEditor(props: {
           <Select size="sm" value={props.rule.status ?? 'active'} onChange={(event) => props.onChange({ ...props.rule, status: event.target.value })}>{statuses.map((item) => <option key={item} value={item}>{item}</option>)}</Select>
         </PricingFormRow>
         <PricingFormRow label="输入单价/1K tokens">
-          <Input size="sm" min="0" step="0.0001" type="number" value={prices.inputTokenPrice} onChange={(event) => updatePrices(Number(event.target.value), prices.outputTokenPrice)} />
+          <Input
+            size="sm"
+            min="0"
+            step="0.0001"
+            type="number"
+            value={prices.inputTokenPrice}
+            onChange={(event) =>
+              updatePrices(
+                Number(event.target.value),
+                prices.cachedInputTokenPrice,
+                prices.outputTokenPrice,
+              )
+            }
+          />
+        </PricingFormRow>
+        <PricingFormRow label="缓存输入单价/1K tokens">
+          <Input
+            size="sm"
+            min="0"
+            step="0.0001"
+            type="number"
+            value={prices.cachedInputTokenPrice}
+            onChange={(event) =>
+              updatePrices(
+                prices.inputTokenPrice,
+                Number(event.target.value),
+                prices.outputTokenPrice,
+              )
+            }
+          />
         </PricingFormRow>
         <PricingFormRow label="输出单价/1K tokens">
-          <Input size="sm" min="0" step="0.0001" type="number" value={prices.outputTokenPrice} onChange={(event) => updatePrices(prices.inputTokenPrice, Number(event.target.value))} />
+          <Input
+            size="sm"
+            min="0"
+            step="0.0001"
+            type="number"
+            value={prices.outputTokenPrice}
+            onChange={(event) =>
+              updatePrices(
+                prices.inputTokenPrice,
+                prices.cachedInputTokenPrice,
+                Number(event.target.value),
+              )
+            }
+          />
         </PricingFormRow>
         <PricingReadonlyRow label="计价单位" value={unitLabel(props.rule.unit)} />
         <PricingReadonlyRow label="计算方式" value={calculatorLabel(props.rule.calculatorType)} />
@@ -462,14 +515,38 @@ function mergeTemplateRules(templates: PricingRuleInput[], rules: PricingRuleInp
 }
 
 function mergeTextRules(rules: PricingRuleInput[], currency: string): PricingRuleInput {
-  const totalRule = rules.find((rule) => rule.resourceType === 'text_total');
+  const totalRule = rules.find((rule) => rule.resourceType === 'text_total' || rule.resourceType === 'text' || rule.ruleKey === 'text');
   const inputRule = rules.find((rule) => rule.resourceType === 'text_input');
+  const cachedInputRule = rules.find(
+    (rule) => rule.resourceType === 'text_cached_input',
+  );
   const outputRule = rules.find((rule) => rule.resourceType === 'text_output');
-  const source = totalRule ?? inputRule ?? outputRule ?? createTextRule(currency, 0.01, 0.03);
+  const source =
+    totalRule ??
+    inputRule ??
+    cachedInputRule ??
+    outputRule ??
+    createTextRule(currency, 0.01, 0.01, 0.03);
   return createTextRule(
     source.currency ?? currency,
-    Number(source.formulaConfig?.inputTokenPrice ?? inputRule?.basePrice ?? source.basePrice ?? 0.01),
+    Number(
-    Number(source.formulaConfig?.outputTokenPrice ?? outputRule?.basePrice ?? 0.03),
+      source.formulaConfig?.inputTokenPrice ??
+        inputRule?.basePrice ??
+        source.basePrice ??
+        0.01,
+    ),
+    Number(
+      source.formulaConfig?.cachedInputTokenPrice ??
+        source.formulaConfig?.inputCacheHitTokenPrice ??
+        cachedInputRule?.basePrice ??
+        source.formulaConfig?.inputTokenPrice ??
+        inputRule?.basePrice ??
+        source.basePrice ??
+        0.01,
+    ),
+    Number(
+      source.formulaConfig?.outputTokenPrice ?? outputRule?.basePrice ?? 0.03,
+    ),
     source,
   );
 }
@@ -478,9 +555,16 @@ function readGroup(value: RecordValue | undefined, group: ModeDefinition['parame
   return isPlainObject(value?.[group.key]) ? value?.[group.key] as RecordValue : group.defaults;
 }
 
-const textFormula = 'input_tokens / 1000 * input_token_price + output_tokens / 1000 * output_token_price';
+const textFormula =
+  'uncached_input_tokens / 1000 * input_token_price + cached_input_tokens / 1000 * cached_input_token_price + output_tokens / 1000 * output_token_price';
 
-function createTextRule(currency: string, inputTokenPrice: number, outputTokenPrice: number, source?: PricingRuleInput): PricingRuleInput {
+function createTextRule(
+  currency: string,
+  inputTokenPrice: number,
+  cachedInputTokenPrice: number,
+  outputTokenPrice: number,
+  source?: PricingRuleInput,
+): PricingRuleInput {
   return {
     ...(source ?? {}),
     ruleKey: 'text',
@@ -493,7 +577,12 @@ function createTextRule(currency: string, inputTokenPrice: number, outputTokenPr
     baseWeight: {},
     dynamicWeight: {},
     dimensionSchema: {
-      metrics: ['input_tokens', 'output_tokens'],
+      metrics: [
+        'input_tokens',
+        'uncached_input_tokens',
+        'cached_input_tokens',
+        'output_tokens',
+      ],
       unitScale: 1000,
       ...(source?.dimensionSchema ?? {}),
     },
@@ -501,6 +590,7 @@ function createTextRule(currency: string, inputTokenPrice: number, outputTokenPr
       ...(source?.formulaConfig ?? {}),
       formula: textFormula,
       inputTokenPrice,
+      cachedInputTokenPrice,
       outputTokenPrice,
     },
     priority: source?.priority ?? 100,
@@ -510,12 +600,36 @@ function createTextRule(currency: string, inputTokenPrice: number, outputTokenPr
 }
 
 function textPrices(rule: PricingRuleInput) {
+  const compatRule = rule as PricingRuleInput & {
+    cachedInputTokenPrice?: number;
+    inputCacheHitTokenPrice?: number;
+    inputTokenPrice?: number;
+    outputTokenPrice?: number;
+  };
+  const inputTokenPrice = Number(
+    rule.formulaConfig?.inputTokenPrice ?? compatRule.inputTokenPrice ?? rule.basePrice ?? 0.01,
+  );
   return {
-    inputTokenPrice: Number(rule.formulaConfig?.inputTokenPrice ?? rule.basePrice ?? 0.01),
+    inputTokenPrice,
-    outputTokenPrice: Number(rule.formulaConfig?.outputTokenPrice ?? 0.03),
+    cachedInputTokenPrice: Number(
+      rule.formulaConfig?.cachedInputTokenPrice ??
+        rule.formulaConfig?.inputCacheHitTokenPrice ??
+        compatRule.cachedInputTokenPrice ??
+        compatRule.inputCacheHitTokenPrice ??
+        inputTokenPrice,
+    ),
+    outputTokenPrice: Number(rule.formulaConfig?.outputTokenPrice ?? compatRule.outputTokenPrice ?? 0.03),
   };
 }
 
+function isTextPricingRule(rule: PricingRuleInput) {
+  const resourceType = String(rule.resourceType ?? '');
+  const ruleKey = String(rule.ruleKey ?? '');
+  return textPricingKeys.has(resourceType) || textPricingKeys.has(ruleKey);
+}
+
+const textPricingKeys = new Set(['text', 'text_total', 'text_input', 'text_cached_input', 'text_output']);
+
 function createRule(ruleKey: string, displayName: string, resourceType: string, unit: string, basePrice: number, currency: string, calculatorType: string, formula: string, dynamicWeight: RecordValue, dimensionSchema: RecordValue): PricingRuleInput {
   return {
     ruleKey,

apps/web/src/pages/admin/PricingRulesPanel.tsx

@@ -266,14 +266,27 @@ function pricingRuleSummaries(ruleSet: PricingRuleSet): PricingRuleSummary[] {
   const rules = ruleSet.rules ?? [];
   const textTotal = rules.find((rule) => rule.resourceType === 'text_total' || rule.ruleKey === 'text');
   const textInput = rules.find((rule) => rule.resourceType === 'text_input');
+  const textCachedInput = rules.find(
+    (rule) => rule.resourceType === 'text_cached_input',
+  );
   const textOutput = rules.find((rule) => rule.resourceType === 'text_output');
   const image = rules.find((rule) => rule.resourceType === 'image');
   const video = rules.find((rule) => rule.resourceType === 'video');
   const items: PricingRuleSummary[] = [];
-  if (textTotal || textInput || textOutput) {
+  if (textTotal || textInput || textCachedInput || textOutput) {
     const inputPrice = Number(textTotal?.formulaConfig?.inputTokenPrice ?? textInput?.basePrice ?? textTotal?.basePrice ?? 0);
+    const cachedInputPrice = Number(
+      textTotal?.formulaConfig?.cachedInputTokenPrice ??
+        textTotal?.formulaConfig?.inputCacheHitTokenPrice ??
+        textCachedInput?.basePrice ??
+        inputPrice,
+    );
     const outputPrice = Number(textTotal?.formulaConfig?.outputTokenPrice ?? textOutput?.basePrice ?? 0);
-    items.push({ kind: 'text', label: '文本', value: `输入 ${formatPrice(inputPrice)} / 输出 ${formatPrice(outputPrice)} / 1K Token` });
+    items.push({
+      kind: 'text',
+      label: '文本',
+      value: `输入 ${formatPrice(inputPrice)} / 缓存输入 ${formatPrice(cachedInputPrice)} / 输出 ${formatPrice(outputPrice)} / 1K Token`,
+    });
   }
   if (image) items.push({ kind: 'image', label: '图像', value: `${formatPrice(image.basePrice)} / ${unitLabel(image.unit)}` });
   if (video) items.push({ kind: 'video', label: '视频', value: `${formatPrice(video.basePrice)} / ${unitLabel(video.unit)}` });