fix(openai): avoid duplicate responses stream text and normalize mini reasoning

core/relay/adaptor/openai/chat.go

@@ -186,9 +186,9 @@ func (s *chatCompletionStreamState) handleFunctionCallArgumentsDelta(
 // handleOutputItemDone handles response.output_item.done event for ChatCompletion
 func (s *chatCompletionStreamState) handleOutputItemDone(
 	event *relaymodel.ResponseStreamEvent,
-) *relaymodel.ChatCompletionsStreamResponse {
+) {
 	if event.Item == nil {
-		return nil
+		return
 	}
 
 	// Handle function call completion
@@ -205,32 +205,8 @@ func (s *chatCompletionStreamState) handleOutputItemDone(
 		s.toolCallArgs = ""
 
 		// No need to send another chunk - arguments already streamed
-		return nil
-	}
-
-	// Handle message content
-	if len(event.Item.Content) > 0 {
-		for _, content := range event.Item.Content {
-			if (content.Type == "text" || content.Type == "output_text") && content.Text != "" {
-				return &relaymodel.ChatCompletionsStreamResponse{
-					ID:      s.messageID,
-					Object:  relaymodel.ChatCompletionChunkObject,
-					Created: time.Now().Unix(),
-					Model:   s.meta.ActualModel,
-					Choices: []*relaymodel.ChatCompletionsStreamResponseChoice{
-						{
-							Index: 0,
-							Delta: relaymodel.Message{
-								Content: content.Text,
-							},
-						},
-					},
-				}
-			}
-		}
+		return
 	}
-
-	return nil
 }
 
 // handleResponseCompleted handles response.completed/done event for ChatCompletion
@@ -1221,7 +1197,7 @@ func ConvertResponsesToChatCompletionStreamResponse(
 		case relaymodel.EventFunctionCallArgumentsDelta:
 			chatStreamResp = state.handleFunctionCallArgumentsDelta(&event)
 		case relaymodel.EventOutputItemDone:
-			chatStreamResp = state.handleOutputItemDone(&event)
+			state.handleOutputItemDone(&event)
 		case relaymodel.EventResponseCompleted, relaymodel.EventResponseDone:
 			chatStreamResp = state.handleResponseCompleted(&event)
 		}

core/relay/adaptor/openai/chat_test.go

@@ -6,6 +6,7 @@ import (
 	"encoding/json"
 	"net/http"
 	"net/http/httptest"
+	"strings"
 	"testing"
 
 	"github.com/gin-gonic/gin"
@@ -515,6 +516,81 @@ func TestConvertResponsesToChatCompletionResponse(t *testing.T) {
 	}
 }
 
+func TestConvertResponsesToChatCompletionStreamResponseSkipsOutputItemDoneContent(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	stream := strings.Join([]string{
+		`data: {"type":"response.created","response":{"id":"resp_123","object":"response","created_at":1780731105,"status":"in_progress","model":"gpt-5.1","output":[],"parallel_tool_calls":true,"store":false}}`,
+		"",
+		`data: {"type":"response.output_item.added","item":{"id":"msg_123","type":"message","role":"assistant","content":[]}}`,
+		"",
+		`data: {"type":"response.output_text.delta","item_id":"msg_123","output_index":0,"content_index":0,"delta":"Hello! What would you like to discuss or work on?"}`,
+		"",
+		`data: {"type":"response.output_item.done","item":{"id":"msg_123","type":"message","role":"assistant","content":[{"type":"output_text","text":"Hello! What would you like to discuss or work on?"}]}}`,
+		"",
+		`data: {"type":"response.completed","response":{"id":"resp_123","object":"response","created_at":1780731105,"status":"completed","model":"gpt-5.1","output":[{"id":"msg_123","type":"message","role":"assistant","content":[{"type":"output_text","text":"Hello! What would you like to discuss or work on?"}]}],"parallel_tool_calls":true,"store":false,"usage":{"input_tokens":7,"output_tokens":22,"total_tokens":29}}}`,
+		"",
+		`data: [DONE]`,
+		"",
+	}, "\n")
+
+	httpResp := &http.Response{
+		StatusCode: http.StatusOK,
+		Body:       &mockReadCloser{Reader: bytes.NewReader([]byte(stream))},
+		Header:     make(http.Header),
+	}
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Request = httptest.NewRequestWithContext(
+		t.Context(),
+		http.MethodPost,
+		"/v1/chat/completions",
+		nil,
+	)
+
+	m := &meta.Meta{
+		ActualModel: "gpt-5.1",
+	}
+
+	_, err := openai.ConvertResponsesToChatCompletionStreamResponse(m, c, httpResp)
+	require.Nil(t, err)
+
+	content := collectChatCompletionStreamContent(t, w.Body.String())
+	assert.Equal(t, "Hello! What would you like to discuss or work on?", content)
+	assert.Equal(
+		t,
+		1,
+		strings.Count(w.Body.String(), "Hello! What would you like to discuss or work on?"),
+	)
+}
+
+func collectChatCompletionStreamContent(t *testing.T, body string) string {
+	t.Helper()
+
+	var builder strings.Builder
+
+	for line := range strings.SplitSeq(body, "\n") {
+		line = strings.TrimSpace(line)
+		if !strings.HasPrefix(line, "data: ") || line == "data: [DONE]" {
+			continue
+		}
+
+		var chunk relaymodel.ChatCompletionsStreamResponse
+
+		err := json.Unmarshal([]byte(strings.TrimPrefix(line, "data: ")), &chunk)
+		require.NoError(t, err)
+
+		for _, choice := range chunk.Choices {
+			if content, ok := choice.Delta.Content.(string); ok {
+				builder.WriteString(content)
+			}
+		}
+	}
+
+	return builder.String()
+}
+
 // mockReadCloser is a helper to create a ReadCloser from a Reader
 type mockReadCloser struct {
 	*bytes.Reader

core/relay/adaptor/openai/gemini_test.go

@@ -61,6 +61,20 @@ func TestConvertGeminiRequest_MapsThinkingConfigToReasoningEffort(t *testing.T)
 			}`,
 			expectedEffort: "low",
 		},
+		{
+			name:        "gpt-5.4 mini snapshot does not receive minimal",
+			actualModel: "gpt-5.4-mini-2026-03-17",
+			requestJSON: `{
+				"generationConfig": {
+					"thinkingConfig": {
+						"thinkingBudget": 512,
+						"includeThoughts": true
+					}
+				},
+				"contents": [{"role":"user","parts":[{"text":"hello"}]}]
+			}`,
+			expectedEffort: "low",
+		},
 		{
 			name:        "gpt-5 does not receive xhigh",
 			actualModel: "gpt-5",

core/relay/adaptor/openai/reasoning.go

@@ -332,7 +332,19 @@ func isKnownOpenAIModelSuffix(suffix string) bool {
 	}
 
 	if matched, ok := strings.CutPrefix(suffix, "-"); ok {
-		return isDateSuffix(matched)
+		if isDateSuffix(matched) {
+			return true
+		}
+
+		for _, variant := range []string{"mini", "nano", "chat-latest"} {
+			if matched == variant {
+				return true
+			}
+
+			if dateSuffix, ok := strings.CutPrefix(matched, variant+"-"); ok {
+				return isDateSuffix(dateSuffix)
+			}
+		}
 	}
 
 	return false

core/relay/adaptor/openai/reasoning_test.go

@@ -127,6 +127,18 @@ func TestOpenAIReasoningEffortForModel(t *testing.T) {
 			effort:      relaymodel.ReasoningEffortMinimal,
 			want:        relaymodel.ReasoningEffortLow,
 		},
+		{
+			name:        "series keyword matching handles mini dated snapshot names",
+			originModel: "gpt-5.4-mini-2026-03-17",
+			effort:      relaymodel.ReasoningEffortMinimal,
+			want:        relaymodel.ReasoningEffortLow,
+		},
+		{
+			name:        "series keyword matching handles nano dated snapshot names",
+			originModel: "gpt-5.4-nano-2026-03-17",
+			effort:      relaymodel.ReasoningEffortMinimal,
+			want:        relaymodel.ReasoningEffortLow,
+		},
 	}
 
 	for _, tt := range tests {
@@ -171,6 +183,13 @@ func TestConvertRequest_OpenAIReasoningEffortCompatibility(t *testing.T) {
 			body:        `{"model":"alias","input":"hi","reasoning":{"effort":"xhigh"}}`,
 			wantEffort:  "high",
 		},
+		{
+			name:        "native responses gpt-5.4 mini snapshot minimal to low",
+			mode:        mode.Responses,
+			actualModel: "gpt-5.4-mini-2026-03-17",
+			body:        `{"model":"alias","input":"hi","reasoning":{"effort":"minimal"}}`,
+			wantEffort:  "low",
+		},
 		{
 			name:        "native chat origin match beats actual fallback",
 			mode:        mode.ChatCompletions,