From 08e078fc255142445728a5a0ca8908f2de630807 Mon Sep 17 00:00:00 2001
From: Ben Vargas <ben@vargas.com>
Date: Thu, 19 Feb 2026 23:25:34 -0700
Subject: [PATCH] fix(openai): route copilot Gemini preview models to chat
 endpoint

---
 internal/registry/model_definitions.go        |  3 ++
 internal/registry/model_definitions_test.go   | 28 ++++++++++++++++++
 sdk/api/handlers/openai/endpoint_compat.go    | 11 ++++++-
 .../handlers/openai/endpoint_compat_test.go   | 29 +++++++++++++++++++
 4 files changed, 70 insertions(+), 1 deletion(-)
 create mode 100644 internal/registry/model_definitions_test.go
 create mode 100644 sdk/api/handlers/openai/endpoint_compat_test.go

diff --git a/internal/registry/model_definitions.go b/internal/registry/model_definitions.go
index 9299ca22..8bab7710 100644
--- a/internal/registry/model_definitions.go
+++ b/internal/registry/model_definitions.go
@@ -571,6 +571,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "Google Gemini 2.5 Pro via GitHub Copilot",
 			ContextLength:       1048576,
 			MaxCompletionTokens: 65536,
+			SupportedEndpoints:  []string{"/chat/completions"},
 		},
 		{
 			ID:                  "gemini-3-pro-preview",
@@ -582,6 +583,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "Google Gemini 3 Pro Preview via GitHub Copilot",
 			ContextLength:       1048576,
 			MaxCompletionTokens: 65536,
+			SupportedEndpoints:  []string{"/chat/completions"},
 		},
 		{
 			ID:                  "gemini-3.1-pro-preview",
@@ -604,6 +606,7 @@ func GetGitHubCopilotModels() []*ModelInfo {
 			Description:         "Google Gemini 3 Flash Preview via GitHub Copilot",
 			ContextLength:       1048576,
 			MaxCompletionTokens: 65536,
+			SupportedEndpoints:  []string{"/chat/completions"},
 		},
 		{
 			ID:                  "grok-code-fast-1",
diff --git a/internal/registry/model_definitions_test.go b/internal/registry/model_definitions_test.go
new file mode 100644
index 00000000..f38b9ce4
--- /dev/null
+++ b/internal/registry/model_definitions_test.go
@@ -0,0 +1,28 @@
+package registry
+
+import "testing"
+
+func TestGitHubCopilotGeminiModelsAreChatOnly(t *testing.T) {
+	models := GetGitHubCopilotModels()
+	required := map[string]bool{
+		"gemini-2.5-pro":         false,
+		"gemini-3-pro-preview":   false,
+		"gemini-3-flash-preview": false,
+	}
+
+	for _, model := range models {
+		if _, ok := required[model.ID]; !ok {
+			continue
+		}
+		required[model.ID] = true
+		if len(model.SupportedEndpoints) != 1 || model.SupportedEndpoints[0] != "/chat/completions" {
+			t.Fatalf("model %q supported endpoints = %v, want [/chat/completions]", model.ID, model.SupportedEndpoints)
+		}
+	}
+
+	for modelID, found := range required {
+		if !found {
+			t.Fatalf("expected GitHub Copilot model %q in definitions", modelID)
+		}
+	}
+}
diff --git a/sdk/api/handlers/openai/endpoint_compat.go b/sdk/api/handlers/openai/endpoint_compat.go
index d7fc5f2f..991e863c 100644
--- a/sdk/api/handlers/openai/endpoint_compat.go
+++ b/sdk/api/handlers/openai/endpoint_compat.go
@@ -1,6 +1,9 @@
 package openai
 
-import "github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+import (
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
+)
 
 const (
 	openAIChatEndpoint      = "/chat/completions"
@@ -12,6 +15,12 @@ func resolveEndpointOverride(modelName, requestedEndpoint string) (string, bool)
 		return "", false
 	}
 	info := registry.GetGlobalRegistry().GetModelInfo(modelName, "")
+	if info == nil {
+		baseModel := thinking.ParseSuffix(modelName).ModelName
+		if baseModel != "" && baseModel != modelName {
+			info = registry.GetGlobalRegistry().GetModelInfo(baseModel, "")
+		}
+	}
 	if info == nil || len(info.SupportedEndpoints) == 0 {
 		return "", false
 	}
diff --git a/sdk/api/handlers/openai/endpoint_compat_test.go b/sdk/api/handlers/openai/endpoint_compat_test.go
new file mode 100644
index 00000000..b622e6dd
--- /dev/null
+++ b/sdk/api/handlers/openai/endpoint_compat_test.go
@@ -0,0 +1,29 @@
+package openai
+
+import (
+	"testing"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+)
+
+func TestResolveEndpointOverride_StripsThinkingSuffix(t *testing.T) {
+	const clientID = "test-endpoint-compat-suffix"
+	reg := registry.GetGlobalRegistry()
+	reg.RegisterClient(clientID, "github-copilot", []*registry.ModelInfo{
+		{
+			ID:                 "test-gemini-chat-only",
+			SupportedEndpoints: []string{openAIChatEndpoint},
+		},
+	})
+	t.Cleanup(func() {
+		reg.UnregisterClient(clientID)
+	})
+
+	override, ok := resolveEndpointOverride("test-gemini-chat-only(high)", openAIResponsesEndpoint)
+	if !ok {
+		t.Fatalf("expected endpoint override to be resolved")
+	}
+	if override != openAIChatEndpoint {
+		t.Fatalf("override endpoint = %q, want %q", override, openAIChatEndpoint)
+	}
+}