fix: preserve Qwen thinking_budget passthrough (#5836)

* fix: preserve qwen thinking budget

* test: address qwen thinking budget review comments

* chore: remove unreachable adaptor code

* test: cover zero Qwen thinking budgets

relaykit/relayconvert/internal/oai_chat/to_oai_responses_req.go

@@ -386,6 +386,8 @@ func ChatCompletionsRequestToResponsesRequest(req *dto.GeneralOpenAIRequest) (*d
 		ParallelToolCalls: parallelToolCallsRaw,
 		Store:             req.Store,
 		Metadata:          req.Metadata,
+		EnableThinking:    req.EnableThinking,
+		ThinkingBudget:    req.ThinkingBudget,
 	}
 	if req.MaxTokens != nil || req.MaxCompletionTokens != nil {
 		out.MaxOutputTokens = lo.ToPtr(maxOutputTokens)

relaykit/relayconvert/internal/oai_chat/to_oai_responses_req_test.go

@@ -1,9 +1,11 @@
 package oaichat
 
 import (
+	"encoding/json"
 	"testing"
 
 	"github.com/QuantumNous/new-api/relaykit/dto"
+	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
 	"github.com/samber/lo"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -37,6 +39,42 @@ func TestChatCompletionsRequestToResponsesRequestInstructionsAndTools(t *testing
 	assert.Equal(t, "function_call_output", gjson.GetBytes(got.Input, "3.type").String())
 }
 
+func TestChatCompletionsRequestToResponsesRequestPreservesQwenThinkingBudget(t *testing.T) {
+	tests := []struct {
+		name   string
+		budget json.RawMessage
+		want   int64
+	}{
+		{name: "positive budget", budget: json.RawMessage(`128`), want: 128},
+		{name: "zero budget", budget: json.RawMessage(`0`), want: 0},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			req := &dto.GeneralOpenAIRequest{
+				Model:          "qwen-plus",
+				EnableThinking: json.RawMessage(`true`),
+				ThinkingBudget: tt.budget,
+				Messages: []dto.Message{
+					{Role: "user", Content: "hello"},
+				},
+			}
+
+			got, err := ChatCompletionsRequestToResponsesRequest(req)
+			require.NoError(t, err)
+			assert.Equal(t, tt.budget, got.ThinkingBudget)
+
+			encoded, err := kitutil.Marshal(got)
+			require.NoError(t, err)
+
+			assert.True(t, gjson.GetBytes(encoded, "enable_thinking").Bool())
+			value := gjson.GetBytes(encoded, "thinking_budget")
+			assert.True(t, value.Exists())
+			assert.Equal(t, tt.want, value.Int())
+		})
+	}
+}
+
 func TestChatCompletionsRequestToResponsesRequestRejectsMultipleChoices(t *testing.T) {
 	_, err := ChatCompletionsRequestToResponsesRequest(&dto.GeneralOpenAIRequest{
 		Model: "gpt-test",

relaykit/relayconvert/internal/oai_responses/to_oai_chat_req.go

@@ -73,6 +73,7 @@ func ResponsesRequestToChatCompletionsRequest(req *dto.OpenAIResponsesRequest) (
 		SafetyIdentifier:     req.SafetyIdentifier,
 		PromptCacheRetention: req.PromptCacheRetention,
 		EnableThinking:       req.EnableThinking,
+		ThinkingBudget:       req.ThinkingBudget,
 	}
 
 	if req.Reasoning != nil {

relaykit/relayconvert/internal/oai_responses/to_oai_chat_req_test.go

@@ -1,6 +1,7 @@
 package oairesponses
 
 import (
+	"encoding/json"
 	"testing"
 
 	"github.com/QuantumNous/new-api/relaykit/dto"
@@ -55,6 +56,38 @@ func TestResponsesRequestToChatCompletionsRequestInstructionsAndScalarInput(t *t
 	assert.Equal(t, "abc", gjson.GetBytes(got.Metadata, "trace").String())
 }
 
+func TestResponsesRequestToChatCompletionsRequestPreservesQwenThinkingBudget(t *testing.T) {
+	tests := []struct {
+		name   string
+		budget json.RawMessage
+		want   int64
+	}{
+		{name: "positive budget", budget: json.RawMessage(`128`), want: 128},
+		{name: "zero budget", budget: json.RawMessage(`0`), want: 0},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got, err := ResponsesRequestToChatCompletionsRequest(&dto.OpenAIResponsesRequest{
+				Model:          "qwen-plus",
+				Input:          mustRawMessage(t, "hello"),
+				EnableThinking: json.RawMessage(`true`),
+				ThinkingBudget: tt.budget,
+			})
+			require.NoError(t, err)
+			assert.Equal(t, tt.budget, got.ThinkingBudget)
+
+			encoded, err := kitutil.Marshal(got)
+			require.NoError(t, err)
+
+			assert.True(t, gjson.GetBytes(encoded, "enable_thinking").Bool())
+			value := gjson.GetBytes(encoded, "thinking_budget")
+			assert.True(t, value.Exists())
+			assert.Equal(t, tt.want, value.Int())
+		})
+	}
+}
+
 func TestResponsesRequestToChatCompletionsRequestMultimodalInput(t *testing.T) {
 	got, err := ResponsesRequestToChatCompletionsRequest(&dto.OpenAIResponsesRequest{
 		Model: "gpt-test",