diff --git a/internal/translator/gemini_helper.go b/internal/translator/gemini_helper.go
index e126bfe4a4..9d9178f2b5 100644
--- a/internal/translator/gemini_helper.go
+++ b/internal/translator/gemini_helper.go
@@ -603,12 +603,21 @@ func isGeminiFlashModel(model internalapi.RequestModel) bool {
 	return strings.Contains(strings.ToLower(model), "flash")
 }
 
+// supportsMediumThinkingLevel reports whether the model accepts the "medium" thinking level.
+// Gemini 3 Pro only lists "low" and "high", and Gemini 3.1 Flash-Lite Image only "minimal" and "high";
+// Flash and Gemini 3.1 Pro also list "medium".
+// https://ai.google.dev/gemini-api/docs/thinking#thinking-levels
+func supportsMediumThinkingLevel(model internalapi.RequestModel) bool {
+	m := strings.ToLower(model)
+	return !strings.Contains(m, "gemini-3-pro") && !strings.Contains(m, "flash-lite-image")
+}
+
 // mapReasoningEffortToThinkingLevel converts OpenAI reasoning effort levels to Gemini thinking levels.
 // The mapping depends on the model type:
 // - "none" → ThinkingLevelMinimal (Gemini Flash only)
 // - "low" → ThinkingLevelLow
-// - "medium" → ThinkingLevelMedium for Flash, ThinkingLevelHigh for Pro
-// - "high" → ThinkingLevelHigh
+// - "medium" → ThinkingLevelMedium, except ThinkingLevelHigh for Gemini 3 Pro and 3.1 Flash-Lite Image, which have no medium level
+// - "high" → ThinkingLevelHigh (every Gemini 3 model; it is the Pro default)
 // https://docs.cloud.google.com/vertex-ai/generative-ai/docs/start/get-started-with-gemini-3#openai-example
 func mapReasoningEffortToThinkingLevel(reasonEffort openai.ReasoningEffort, model internalapi.RequestModel) (genai.ThinkingLevel, error) {
 	isFlash := isGeminiFlashModel(model)
@@ -622,14 +631,11 @@ func mapReasoningEffortToThinkingLevel(reasonEffort openai.ReasoningEffort, mode
 	case openai.ReasoningEffortLow:
 		return genai.ThinkingLevelLow, nil
 	case openai.ReasoningEffortMedium:
-		if isFlash {
+		if supportsMediumThinkingLevel(model) {
 			return genai.ThinkingLevelMedium, nil
 		}
 		return genai.ThinkingLevelHigh, nil
 	case openai.ReasoningEffortHigh:
-		if !isFlash {
-			return "", fmt.Errorf("%w: reasoning effort 'high' is only supported for Gemini Flash models", internalapi.ErrInvalidRequestBody)
-		}
 		return genai.ThinkingLevelHigh, nil
 	default:
 		return "", fmt.Errorf("%w: unsupported reasoning effort level: %q (supported: none, low, medium, high)", internalapi.ErrInvalidRequestBody, reasonEffort)
diff --git a/internal/translator/gemini_helper_test.go b/internal/translator/gemini_helper_test.go
index 529efc726a..b7e3f3bd1e 100644
--- a/internal/translator/gemini_helper_test.go
+++ b/internal/translator/gemini_helper_test.go
@@ -1577,6 +1577,19 @@ func TestOpenAIReqToGeminiGenerationConfig(t *testing.T) {
 			expectedResponseMode: responseModeNone,
 			requestModel:         "gemini-3-flash",
 		},
+		{
+			name: "high reasoning effort on Pro maps to ThinkingLevelHigh",
+			input: &openai.ChatCompletionRequest{
+				ReasoningEffort: openai.ReasoningEffortHigh,
+			},
+			expectedGenerationConfig: &genai.GenerationConfig{
+				ThinkingConfig: &genai.ThinkingConfig{
+					ThinkingLevel: genai.ThinkingLevelHigh,
+				},
+			},
+			expectedResponseMode: responseModeNone,
+			requestModel:         "gemini-3-pro",
+		},
 		{
 			name: "reasoning effort unsupported value",
 			input: &openai.ChatCompletionRequest{
@@ -3448,17 +3461,41 @@ func TestMapReasoningEffortToThinkingLevel(t *testing.T) {
 			expectedThinking: genai.ThinkingLevelMedium,
 		},
 		{
-			name:             "medium effort on Pro maps to ThinkingLevelHigh",
+			name:             "medium effort on Gemini 3 Pro maps to ThinkingLevelHigh",
 			reasoningEffort:  openai.ReasoningEffortMedium,
 			model:            "gemini-3-pro",
 			expectedThinking: genai.ThinkingLevelHigh,
 		},
 		{
-			name:             "high effort maps to ThinkingLevelHigh",
+			name:             "medium effort on Gemini 3 Pro preview maps to ThinkingLevelHigh",
+			reasoningEffort:  openai.ReasoningEffortMedium,
+			model:            "gemini-3-pro-preview",
+			expectedThinking: genai.ThinkingLevelHigh,
+		},
+		{
+			name:             "medium effort on Gemini 3.1 Pro preview maps to ThinkingLevelMedium",
+			reasoningEffort:  openai.ReasoningEffortMedium,
+			model:            "gemini-3.1-pro-preview",
+			expectedThinking: genai.ThinkingLevelMedium,
+		},
+		{
+			name:             "medium effort on Gemini 3.1 Flash-Lite Image maps to ThinkingLevelHigh",
+			reasoningEffort:  openai.ReasoningEffortMedium,
+			model:            "gemini-3.1-flash-lite-image",
+			expectedThinking: genai.ThinkingLevelHigh,
+		},
+		{
+			name:             "high effort on Flash maps to ThinkingLevelHigh",
 			reasoningEffort:  openai.ReasoningEffortHigh,
 			model:            "gemini-3-flash",
 			expectedThinking: genai.ThinkingLevelHigh,
 		},
+		{
+			name:             "high effort on Pro maps to ThinkingLevelHigh",
+			reasoningEffort:  openai.ReasoningEffortHigh,
+			model:            "gemini-3-pro",
+			expectedThinking: genai.ThinkingLevelHigh,
+		},
 		{
 			name:             "empty effort - not supported",
 			reasoningEffort:  "",