diff --git a/pkg/providers/factory_provider.go b/pkg/providers/factory_provider.go
index 7d5566eef..fba34ed00 100644
--- a/pkg/providers/factory_provider.go
+++ b/pkg/providers/factory_provider.go
@@ -84,7 +84,9 @@ func CreateProviderFromConfig(cfg *config.ModelConfig) (LLMProvider, string, err
 		if apiBase == "" {
 			apiBase = getDefaultAPIBase(protocol)
 		}
-		return NewHTTPProviderWithMaxTokensField(cfg.APIKey, apiBase, cfg.Proxy, cfg.MaxTokensField), modelID, nil
+		p := NewHTTPProviderWithMaxTokensField(cfg.APIKey, apiBase, cfg.Proxy, cfg.MaxTokensField)
+		p.SetSupportPromptCache(true)
+		return p, modelID, nil
 
 	case "openrouter", "groq", "zhipu", "gemini", "nvidia",
 		"ollama", "moonshot", "shengsuanyun", "deepseek", "cerebras",
@@ -97,7 +99,12 @@ func CreateProviderFromConfig(cfg *config.ModelConfig) (LLMProvider, string, err
 		if apiBase == "" {
 			apiBase = getDefaultAPIBase(protocol)
 		}
-		return NewHTTPProviderWithMaxTokensField(cfg.APIKey, apiBase, cfg.Proxy, cfg.MaxTokensField), modelID, nil
+		p := NewHTTPProviderWithMaxTokensField(cfg.APIKey, apiBase, cfg.Proxy, cfg.MaxTokensField)
+		// Gemini rejects unknown fields in the request body; disable prompt_cache_key for it.
+		if protocol != "gemini" {
+			p.SetSupportPromptCache(true)
+		}
+		return p, modelID, nil
 
 	case "anthropic":
 		if cfg.AuthMethod == "oauth" || cfg.AuthMethod == "token" {
diff --git a/pkg/providers/http_provider.go b/pkg/providers/http_provider.go
index d0c4344f3..166a8611e 100644
--- a/pkg/providers/http_provider.go
+++ b/pkg/providers/http_provider.go
@@ -38,6 +38,11 @@ func (p *HTTPProvider) Chat(
 	return p.delegate.Chat(ctx, messages, tools, model, options)
 }
 
+// SetSupportPromptCache enables or disables sending the prompt_cache_key field.
+func (p *HTTPProvider) SetSupportPromptCache(v bool) {
+	p.delegate.SetSupportPromptCache(v)
+}
+
 func (p *HTTPProvider) GetDefaultModel() string {
 	return ""
 }
diff --git a/pkg/providers/openai_compat/provider.go b/pkg/providers/openai_compat/provider.go
index a8d244d4a..387bb50e8 100644
--- a/pkg/providers/openai_compat/provider.go
+++ b/pkg/providers/openai_compat/provider.go
@@ -28,10 +28,11 @@ type (
 )
 
 type Provider struct {
-	apiKey         string
-	apiBase        string
-	maxTokensField string // Field name for max tokens (e.g., "max_completion_tokens" for o1/glm models)
-	httpClient     *http.Client
+	apiKey              string
+	apiBase             string
+	maxTokensField      string // Field name for max tokens (e.g., "max_completion_tokens" for o1/glm models)
+	httpClient          *http.Client
+	supportPromptCache  bool // Only send prompt_cache_key when true (OpenAI-specific feature)
 }
 
 func NewProvider(apiKey, apiBase, proxy string) *Provider {
@@ -62,6 +63,12 @@ func NewProviderWithMaxTokensField(apiKey, apiBase, proxy, maxTokensField string
 	}
 }
 
+// SetSupportPromptCache enables or disables sending the prompt_cache_key field.
+// Most providers accept or ignore this field, but Gemini rejects unknown fields.
+func (p *Provider) SetSupportPromptCache(v bool) {
+	p.supportPromptCache = v
+}
+
 func (p *Provider) Chat(
 	ctx context.Context,
 	messages []Message,
@@ -115,8 +122,11 @@ func (p *Provider) Chat(
 	// with the same key and reuse prefix KV cache across calls.
 	// The key is typically the agent ID — stable per agent, shared across requests.
 	// See: https://platform.openai.com/docs/guides/prompt-caching
-	if cacheKey, ok := options["prompt_cache_key"].(string); ok && cacheKey != "" {
-		requestBody["prompt_cache_key"] = cacheKey
+	// Disabled for providers like Gemini that reject unknown fields in the request body.
+	if p.supportPromptCache {
+		if cacheKey, ok := options["prompt_cache_key"].(string); ok && cacheKey != "" {
+			requestBody["prompt_cache_key"] = cacheKey
+		}
 	}
 
 	jsonData, err := json.Marshal(requestBody)