fix(azure): support GPT-5 and o-series reasoning models

- Update default API version from 2024-05-01-preview to 2025-04-01-preview
  (required for o-series and GPT-5 models)
- Remove NeedsRawMode override that always returned false, inheriting parent
  logic that correctly skips temperature/top_p for reasoning models
- Add /responses route to deployment middleware for future v1 API support

Tested against Azure OpenAI gpt-5.2-chat deployment.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

internal/plugins/ai/azure/azure.go

@@ -35,7 +35,7 @@ type Client struct {
 	apiDeployments []string
 }
 
-const defaultAPIVersion = "2024-05-01-preview"
+const defaultAPIVersion = "2025-04-01-preview"
 
 func (oi *Client) configure() error {
 	oi.apiDeployments = parseDeployments(oi.ApiDeployments.Value)
@@ -86,6 +86,7 @@ func azureDeploymentMiddleware(req *http.Request, next option.MiddlewareNext) (*
 		"/audio/transcriptions": true,
 		"/audio/translations":   true,
 		"/images/generations":   true,
+		"/responses":            true,
 	}
 
 	path := req.URL.Path
@@ -156,6 +157,3 @@ func (oi *Client) ListModels() (ret []string, err error) {
 	return
 }
 
-func (oi *Client) NeedsRawMode(modelName string) bool {
-	return false
-}

internal/plugins/ai/azure/azure_test.go

@@ -1,6 +1,9 @@
 package azure
 
 import (
+	"bytes"
+	"io"
+	"net/http"
 	"testing"
 )
 
@@ -27,7 +30,7 @@ func TestClientConfigure(t *testing.T) {
 	client.ApiDeployments.Value = "deployment1,deployment2"
 	client.ApiKey.Value = "test-api-key"
 	client.ApiBaseURL.Value = "https://example.com"
-	client.ApiVersion.Value = "2024-05-01-preview"
+	client.ApiVersion.Value = "2025-04-01-preview"
 
 	err := client.configure()
 	if err != nil {
@@ -48,8 +51,8 @@ func TestClientConfigure(t *testing.T) {
 		t.Errorf("Expected ApiClient to be initialized, got nil")
 	}
 
-	if client.ApiVersion.Value != "2024-05-01-preview" {
-		t.Errorf("Expected API version to be '2024-05-01-preview', got %s", client.ApiVersion.Value)
+	if client.ApiVersion.Value != "2025-04-01-preview" {
+		t.Errorf("Expected API version to be '2025-04-01-preview', got %s", client.ApiVersion.Value)
 	}
 }
 
@@ -88,3 +91,81 @@ func TestListModels(t *testing.T) {
 		}
 	}
 }
+
+func TestNeedsRawModeInheritsFromParent(t *testing.T) {
+	client := NewClient()
+
+	tests := []struct {
+		name     string
+		model    string
+		expected bool
+	}{
+		{"o1 model", "o1", true},
+		{"o1-preview", "o1-preview", true},
+		{"o3-mini", "o3-mini", true},
+		{"o4-mini", "o4-mini", true},
+		{"gpt-5", "gpt-5", true},
+		{"gpt-5-turbo", "gpt-5-turbo", true},
+		{"gpt-4o", "gpt-4o", false},
+		{"gpt-4", "gpt-4", false},
+		{"regular deployment", "my-deployment", false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := client.NeedsRawMode(tt.model)
+			if result != tt.expected {
+				t.Errorf("NeedsRawMode(%q) = %v, want %v", tt.model, result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestMiddlewareResponsesRoute(t *testing.T) {
+	// Verify /responses is in the deployment routes by testing the middleware
+	body := `{"model": "gpt-5"}`
+	req, err := http.NewRequest("POST", "https://example.com/openai/responses", io.NopCloser(bytes.NewReader([]byte(body))))
+	if err != nil {
+		t.Fatalf("Failed to create request: %v", err)
+	}
+
+	var capturedPath string
+	mockNext := func(req *http.Request) (*http.Response, error) {
+		capturedPath = req.URL.Path
+		return &http.Response{StatusCode: 200}, nil
+	}
+
+	_, err = azureDeploymentMiddleware(req, mockNext)
+	if err != nil {
+		t.Fatalf("Middleware returned error: %v", err)
+	}
+
+	expected := "/openai/deployments/gpt-5/responses"
+	if capturedPath != expected {
+		t.Errorf("Expected path %q, got %q", expected, capturedPath)
+	}
+}
+
+func TestMiddlewareChatCompletionsRoute(t *testing.T) {
+	body := `{"model": "gpt-4o"}`
+	req, err := http.NewRequest("POST", "https://example.com/openai/chat/completions", io.NopCloser(bytes.NewReader([]byte(body))))
+	if err != nil {
+		t.Fatalf("Failed to create request: %v", err)
+	}
+
+	var capturedPath string
+	mockNext := func(req *http.Request) (*http.Response, error) {
+		capturedPath = req.URL.Path
+		return &http.Response{StatusCode: 200}, nil
+	}
+
+	_, err = azureDeploymentMiddleware(req, mockNext)
+	if err != nil {
+		t.Fatalf("Middleware returned error: %v", err)
+	}
+
+	expected := "/openai/deployments/gpt-4o/chat/completions"
+	if capturedPath != expected {
+		t.Errorf("Expected path %q, got %q", expected, capturedPath)
+	}
+}