Preserve exact Gemini high model

codex-luna-compat.patch

@@ -86,6 +86,77 @@ index 0000000..cd731ad
 +	require.Equal(t, "https://example.test/v1beta/models/gemini-3.7-flash:generateContent", ordinaryURL)
 +	require.Equal(t, "gemini-3.7-flash", ordinaryInfo.UpstreamModelName)
 +}
+diff --git a/relay/channel/openai/adaptor.go b/relay/channel/openai/adaptor.go
+index 3fd2466..bd52825 100644
+--- a/relay/channel/openai/adaptor.go
++++ b/relay/channel/openai/adaptor.go
+@@ -602,6 +602,16 @@ func detectImageMimeType(filename string) string {
+ }
+ 
+ func (a *Adaptor) ConvertOpenAIResponsesRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.OpenAIResponsesRequest) (any, error) {
++	if request.Model == codexLunaUpstreamModel ||
++		(info != nil && info.OriginModelName == codexLunaOriginModel) {
++		request.Model = codexLunaUpstreamModel
++		request.Reasoning = nil
++		if info != nil {
++			info.ReasoningEffort = ""
++		}
++		return request, nil
++	}
++
+ 	//  转换模型推理力度后缀
+ 	effort, originModel := reasoning.ParseOpenAIReasoningEffortFromModelSuffix(request.Model)
+ 	if effort != "" {
+diff --git a/relay/channel/openai/adaptor_responses_luna_test.go b/relay/channel/openai/adaptor_responses_luna_test.go
+new file mode 100644
+index 0000000..2d79925
+--- /dev/null
++++ b/relay/channel/openai/adaptor_responses_luna_test.go
+@@ -0,0 +1,44 @@
++package openai
++
++import (
++	"testing"
++
++	"github.com/QuantumNous/new-api/dto"
++	relaycommon "github.com/QuantumNous/new-api/relay/common"
++	"github.com/stretchr/testify/require"
++)
++
++func TestConvertOpenAIResponsesRequestPreservesExactGeminiHighModel(t *testing.T) {
++	info := &relaycommon.RelayInfo{
++		OriginModelName: "gemini-3.7-flash-high",
++		ReasoningEffort: "low",
++	}
++	request := dto.OpenAIResponsesRequest{
++		Model: codexLunaUpstreamModel,
++		Reasoning: &dto.Reasoning{
++			Effort:  "low",
++			Summary: "detailed",
++		},
++	}
++
++	converted, err := (&Adaptor{}).ConvertOpenAIResponsesRequest(nil, info, request)
++	require.NoError(t, err)
++	actual, ok := converted.(dto.OpenAIResponsesRequest)
++	require.True(t, ok)
++	require.Equal(t, codexLunaUpstreamModel, actual.Model)
++	require.Nil(t, actual.Reasoning)
++	require.Empty(t, info.ReasoningEffort)
++}
++
++func TestConvertOpenAIResponsesRequestStillParsesOtherEffortSuffixes(t *testing.T) {
++	info := &relaycommon.RelayInfo{OriginModelName: "gemini-3.7-flash-low"}
++	request := dto.OpenAIResponsesRequest{Model: "gemini-3.7-flash-low"}
++
++	converted, err := (&Adaptor{}).ConvertOpenAIResponsesRequest(nil, info, request)
++	require.NoError(t, err)
++	actual, ok := converted.(dto.OpenAIResponsesRequest)
++	require.True(t, ok)
++	require.Equal(t, "gemini-3.7-flash", actual.Model)
++	require.NotNil(t, actual.Reasoning)
++	require.Equal(t, "low", actual.Reasoning.Effort)
++}
 diff --git a/relay/channel/openai/relay_responses.go b/relay/channel/openai/relay_responses.go
 index 697cc96..299247f 100644
 --- a/relay/channel/openai/relay_responses.go