diff --git a/backend/internal/service/billing_service.go b/backend/internal/service/billing_service.go
index da36ebdaad..74055a8151 100644
--- a/backend/internal/service/billing_service.go
+++ b/backend/internal/service/billing_service.go
@@ -91,6 +91,7 @@ type BillingCache interface {
 type ModelPricing struct {
 	InputPricePerToken             float64 // 每token输入价格 (USD)
 	InputPricePerTokenPriority     float64 // priority service tier 下每token输入价格 (USD)
+	ImageInputPricePerToken        float64 // 图片输入 token 价格 (USD)，用于多模态 embedding 等图文不同价场景；为 0 时回退到 InputPricePerToken
 	OutputPricePerToken            float64 // 每token输出价格 (USD)
 	OutputPricePerTokenPriority    float64 // priority service tier 下每token输出价格 (USD)
 	CacheCreationPricePerToken     float64 // 缓存创建每token价格 (USD)
@@ -137,6 +138,7 @@ func serviceTierCostMultiplier(serviceTier string) float64 {
 // UsageTokens 使用的token数量
 type UsageTokens struct {
 	InputTokens           int
+	ImageInputTokens      int
 	OutputTokens          int
 	CacheCreationTokens   int
 	CacheReadTokens       int
@@ -486,6 +488,17 @@ func (s *BillingService) initFallbackPricing() {
 		CacheReadPricePerToken: 0.03e-6,
 		SupportsCacheBreakdown: false,
 	}
+
+	// ---- 火山方舟 豆包 Embedding（多模态向量化）----
+	// doubao-embedding-vision 图文向量化：上游 usage 回传 prompt_tokens_details.{text_tokens,image_tokens}，
+	// 按量付费官方价 文本 ¥0.7/MTok、图片 ¥1.8/MTok；汇率口径 ÷7.14（与本表其他国产模型一致，¥1≈$0.14）。
+	// embedding 无 output，OutputPricePerToken 置 0。
+	s.fallbackPrices["doubao-embedding-vision"] = &ModelPricing{
+		InputPricePerToken:      0.098e-6, // ¥0.7/MTok ≈ $0.098（文本输入）
+		ImageInputPricePerToken: 0.252e-6, // ¥1.8/MTok ≈ $0.252（图片输入）
+		OutputPricePerToken:     0,
+		SupportsCacheBreakdown:  false,
+	}
 }
 
 // getFallbackPricing 根据模型系列获取回退价格
@@ -621,6 +634,13 @@ func (s *BillingService) getFallbackPricing(model string) *ModelPricing {
 		return s.fallbackPrices["minimax-m2"]
 	}
 
+	// 火山方舟 豆包 Embedding（多模态向量化）。
+	// most-specific-first：放在未来任何 doubao-embedding / doubao 宽匹配之前。
+	// 覆盖带版本后缀的别名（如 doubao-embedding-vision-251215）。
+	if strings.Contains(modelLower, "doubao-embedding-vision") {
+		return s.fallbackPrices["doubao-embedding-vision"]
+	}
+
 	// OpenAI（GPT-5 / Codex 族）：仅匹配已知型号，避免未知 OpenAI 型号误计价。
 	if normalized := normalizeKnownOpenAICodexModel(modelLower); normalized != "" {
 		switch normalized {
@@ -839,7 +859,24 @@ func (s *BillingService) computeTokenBreakdown(
 	}
 
 	bd := &CostBreakdown{}
-	bd.InputCost = float64(tokens.InputTokens) * inputPrice
+	// 分离图片输入 token 与文本输入 token（多模态 embedding 等图文不同价场景）。
+	// ImageInputTokens 为 0 时（绝大多数 chat/vision 流量）走原始单价路径，行为不变。
+	if tokens.ImageInputTokens > 0 {
+		imageInputTokens := tokens.ImageInputTokens
+		textInputTokens := tokens.InputTokens - imageInputTokens
+		if textInputTokens < 0 {
+			textInputTokens = 0
+			imageInputTokens = tokens.InputTokens
+		}
+		imageInputPrice := pricing.ImageInputPricePerToken
+		if imageInputPrice == 0 {
+			// 未配置图片输入档时回退到文本 input 价（已含 priority / 长上下文调整）
+			imageInputPrice = inputPrice
+		}
+		bd.InputCost = float64(textInputTokens)*inputPrice + float64(imageInputTokens)*imageInputPrice
+	} else {
+		bd.InputCost = float64(tokens.InputTokens) * inputPrice
+	}
 
 	// 分离图片输出 token 与文本输出 token
 	textOutputTokens := tokens.OutputTokens - tokens.ImageOutputTokens
diff --git a/backend/internal/service/billing_service_test.go b/backend/internal/service/billing_service_test.go
index 45df3cd6bf..9541047d25 100644
--- a/backend/internal/service/billing_service_test.go
+++ b/backend/internal/service/billing_service_test.go
@@ -582,9 +582,24 @@ func TestGetFallbackPricing_FamilyMatching(t *testing.T) {
 			expectedCacheRead: floatPtr(0.03e-6),
 		},
 
+		// ---- 火山方舟 豆包 Embedding（多模态向量化）----
+		{
+			name:           "doubao embedding vision text rate",
+			model:          "doubao-embedding-vision",
+			expectedInput:  0.098e-6,
+			expectedOutput: floatPtr(0),
+		},
+		{
+			name:          "doubao embedding vision versioned alias",
+			model:         "doubao-embedding-vision-251215",
+			expectedInput: 0.098e-6,
+		},
+
 		// ---- 负向用例 ----
 		{name: "qwen unknown no fallback", model: "qwen-max", expectNilPricing: true},
+		// doubao-pro / doubao-embedding（纯文本）不在白名单，不回退；仅 doubao-embedding-vision 显式命中。
 		{name: "doubao unknown no fallback", model: "doubao-pro", expectNilPricing: true},
+		{name: "doubao text embedding no fallback", model: "doubao-embedding-text-240515", expectNilPricing: true},
 		{name: "hunyuan unknown no fallback", model: "hunyuan-t1", expectNilPricing: true},
 		{name: "moonshot v1 not covered", model: "moonshot-v1-8k", expectNilPricing: true},
 		// kimi-k2-0905 / kimi-k2-0711 官方未公布独立价，走 kimi-k2 隐性回退（接受）——
@@ -618,6 +633,52 @@ func TestGetFallbackPricing_FamilyMatching(t *testing.T) {
 		})
 	}
 }
+
+// doubao-embedding-vision 是首个图文不同价的 embedding：文本 ¥0.7/MTok、图片 ¥1.8/MTok。
+// 验证回退表同时携带文本与图片两档单价，且能被带版本后缀 / 大小写别名命中。
+func TestGetModelPricing_DoubaoEmbeddingVisionImageInputRate(t *testing.T) {
+	svc := newTestBillingService()
+
+	for _, model := range []string{
+		"doubao-embedding-vision",
+		"doubao-embedding-vision-251215",
+		"Doubao-Embedding-Vision",
+	} {
+		pricing, err := svc.GetModelPricing(model)
+		require.NoError(t, err, "model %s should resolve fallback pricing", model)
+		require.NotNil(t, pricing)
+		require.InDelta(t, 0.098e-6, pricing.InputPricePerToken, 1e-12, "text input rate for %s", model)
+		require.InDelta(t, 0.252e-6, pricing.ImageInputPricePerToken, 1e-12, "image input rate for %s", model)
+		require.Zero(t, pricing.OutputPricePerToken, "embedding has no output cost for %s", model)
+	}
+}
+
+// 验证双档计费：InputCost = 文本token×文本价 + 图片token×图片价；
+// 且 ImageInputTokens=0 时走原单价路径，ImageInputTokens>InputTokens 时不负计文本。
+func TestCalculateCost_DoubaoEmbeddingVisionDifferentialInput(t *testing.T) {
+	svc := newTestBillingService()
+
+	// 图文混合：prompt_tokens=1340，其中 image_tokens=28、text_tokens=1312。
+	mixed := UsageTokens{InputTokens: 1340, ImageInputTokens: 28}
+	cost, err := svc.CalculateCost("doubao-embedding-vision", mixed, 1.0)
+	require.NoError(t, err)
+	wantMixed := float64(1312)*0.098e-6 + float64(28)*0.252e-6
+	require.InDelta(t, wantMixed, cost.InputCost, 1e-15)
+	require.InDelta(t, wantMixed, cost.TotalCost, 1e-15)
+	require.Zero(t, cost.OutputCost)
+
+	// 纯文本：全部按文本档计费，与原单价路径一致。
+	textOnly := UsageTokens{InputTokens: 1340}
+	costText, err := svc.CalculateCost("doubao-embedding-vision", textOnly, 1.0)
+	require.NoError(t, err)
+	require.InDelta(t, float64(1340)*0.098e-6, costText.InputCost, 1e-15)
+
+	// 健壮性：ImageInputTokens 超过 InputTokens 时，文本置 0、计费 token 不超过 InputTokens。
+	weird := UsageTokens{InputTokens: 10, ImageInputTokens: 50}
+	costWeird, err := svc.CalculateCost("doubao-embedding-vision", weird, 1.0)
+	require.NoError(t, err)
+	require.InDelta(t, float64(10)*0.252e-6, costWeird.InputCost, 1e-15)
+}
 func TestCalculateCostWithLongContext_BelowThreshold(t *testing.T) {
 	svc := newTestBillingService()
 
diff --git a/backend/internal/service/openai_embeddings.go b/backend/internal/service/openai_embeddings.go
index 7c710259f3..0fb3fff1f7 100644
--- a/backend/internal/service/openai_embeddings.go
+++ b/backend/internal/service/openai_embeddings.go
@@ -214,8 +214,15 @@ func extractOpenAIEmbeddingsUsage(body []byte) OpenAIUsage {
 		usage.Get("cache_creation_input_tokens"),
 		usage.Get("input_tokens_details.cache_creation_tokens"),
 	)
+	// 多模态 embedding（如 doubao-embedding-vision）回传图文 token 拆分，
+	// 用于图文不同价计费；纯文本 embedding 该字段为 0，行为不变。
+	imageInputTokens := firstPositiveGJSONInt(
+		usage.Get("prompt_tokens_details.image_tokens"),
+		usage.Get("input_tokens_details.image_tokens"),
+	)
 	return OpenAIUsage{
 		InputTokens:              inputTokens,
+		ImageInputTokens:         imageInputTokens,
 		OutputTokens:             outputTokens,
 		CacheReadInputTokens:     cacheReadTokens,
 		CacheCreationInputTokens: cacheCreationTokens,
diff --git a/backend/internal/service/openai_gateway_service.go b/backend/internal/service/openai_gateway_service.go
index d056668a48..7fb53e2f55 100644
--- a/backend/internal/service/openai_gateway_service.go
+++ b/backend/internal/service/openai_gateway_service.go
@@ -212,6 +212,7 @@ func (s *OpenAICodexUsageSnapshot) Normalize() *NormalizedCodexLimits {
 // OpenAIUsage represents OpenAI API response usage
 type OpenAIUsage struct {
 	InputTokens              int `json:"input_tokens"`
+	ImageInputTokens         int `json:"image_input_tokens,omitempty"`
 	OutputTokens             int `json:"output_tokens"`
 	CacheCreationInputTokens int `json:"cache_creation_input_tokens,omitempty"`
 	CacheReadInputTokens     int `json:"cache_read_input_tokens,omitempty"`
@@ -5912,6 +5913,7 @@ func (s *OpenAIGatewayService) RecordUsage(ctx context.Context, input *OpenAIRec
 	// Calculate cost
 	tokens := UsageTokens{
 		InputTokens:         actualInputTokens,
+		ImageInputTokens:    result.Usage.ImageInputTokens,
 		OutputTokens:        result.Usage.OutputTokens,
 		CacheCreationTokens: result.Usage.CacheCreationInputTokens,
 		CacheReadTokens:     result.Usage.CacheReadInputTokens,