diff --git a/backend/tests/test_serialization.py b/backend/tests/test_serialization.py
index b707d7143f..9c699f6aff 100644
--- a/backend/tests/test_serialization.py
+++ b/backend/tests/test_serialization.py
@@ -123,6 +123,25 @@ def test_serialize_messages_tuple():
     assert result == [{"key": "v2"}, {"langgraph_node": "agent"}]
 
 
+def test_serialize_messages_tuple_preserves_reasoning_only_chunk():
+    from langchain_core.messages import AIMessageChunk
+
+    from deerflow.runtime.serialization import serialize_messages_tuple
+
+    chunk = AIMessageChunk(
+        content="",
+        id="run-1",
+        additional_kwargs={"reasoning_content": "Thinking through the task."},
+    )
+    metadata = {"langgraph_node": "agent"}
+
+    result = serialize_messages_tuple((chunk, metadata))
+
+    assert result[0]["content"] == ""
+    assert result[0]["additional_kwargs"]["reasoning_content"] == "Thinking through the task."
+    assert result[1] == metadata
+
+
 def test_serialize_messages_tuple_non_dict_metadata():
     from deerflow.runtime.serialization import serialize_messages_tuple
 
diff --git a/frontend/src/core/threads/hooks.ts b/frontend/src/core/threads/hooks.ts
index 249ea366b1..f53780c483 100644
--- a/frontend/src/core/threads/hooks.ts
+++ b/frontend/src/core/threads/hooks.ts
@@ -484,6 +484,7 @@ export function useThreadStream({
           },
           {
             threadId: threadId,
+            streamMode: ["messages-tuple", "values"],
             streamSubgraphs: true,
             streamResumable: true,
             config: {
diff --git a/frontend/tests/unit/core/messages/utils.test.ts b/frontend/tests/unit/core/messages/utils.test.ts
index 24d014c7ef..8635b320ea 100644
--- a/frontend/tests/unit/core/messages/utils.test.ts
+++ b/frontend/tests/unit/core/messages/utils.test.ts
@@ -2,8 +2,10 @@ import type { Message } from "@langchain/langgraph-sdk";
 import { expect, test } from "vitest";
 
 import {
+  extractReasoningContentFromMessage,
   getAssistantTurnUsageMessages,
   getMessageGroups,
+  hasReasoning,
 } from "@/core/messages/utils";
 
 test("aggregates token usage messages once per assistant turn", () => {
@@ -63,3 +65,26 @@ test("aggregates token usage messages once per assistant turn", () => {
     ),
   ).toEqual([null, null, ["ai-1", "ai-2"], null, ["ai-3"]]);
 });
+
+test("recognizes reasoning-only AI messages", () => {
+  const message = {
+    id: "reasoning-1",
+    type: "ai",
+    content: "",
+    additional_kwargs: {
+      reasoning_content: "I need to compare the options first.",
+    },
+  } as Message;
+
+  expect(hasReasoning(message)).toBe(true);
+  expect(extractReasoningContentFromMessage(message)).toBe(
+    "I need to compare the options first.",
+  );
+  expect(getMessageGroups([message])).toEqual([
+    {
+      id: "reasoning-1",
+      type: "assistant:processing",
+      messages: [message],
+    },
+  ]);
+});