diff --git a/src/kimi_cli/config.py b/src/kimi_cli/config.py
index b135958ca..73845b86d 100644
--- a/src/kimi_cli/config.py
+++ b/src/kimi_cli/config.py
@@ -12,6 +12,7 @@
     SecretStr,
     ValidationError,
     field_serializer,
+    field_validator,
     model_validator,
 )
 from tomlkit.exceptions import TOMLKitError
@@ -44,9 +45,19 @@ class LLMProvider(BaseModel):
     """Environment variables to set before creating the provider instance"""
     custom_headers: dict[str, str] | None = None
     """Custom headers to include in API requests"""
+    reasoning_key: str | None = None
+    """Key name for reasoning/thinking content used by the openai_legacy provider (e.g. "reasoning_content")."""
     oauth: OAuthRef | None = None
     """OAuth credential reference (do not store tokens here)."""
 
+    @field_validator("reasoning_key", mode="before")
+    @classmethod
+    def normalize_reasoning_key(cls, v: str | None) -> str | None:
+        if isinstance(v, str):
+            v = v.strip()
+            return v or None
+        return v
+
     @field_serializer("api_key", when_used="json")
     def dump_secret(self, v: SecretStr):
         return v.get_secret_value()
diff --git a/src/kimi_cli/llm.py b/src/kimi_cli/llm.py
index a72644c05..3dcb1680f 100644
--- a/src/kimi_cli/llm.py
+++ b/src/kimi_cli/llm.py
@@ -152,6 +152,7 @@ def create_llm(
                 model=model.model,
                 base_url=provider.base_url,
                 api_key=resolved_api_key,
+                reasoning_key=provider.reasoning_key,
             )
         case "openai_responses":
             from kosong.contrib.chat_provider.openai_responses import OpenAIResponses
diff --git a/tests/core/test_create_llm.py b/tests/core/test_create_llm.py
index 27bb54d4f..02570ab23 100644
--- a/tests/core/test_create_llm.py
+++ b/tests/core/test_create_llm.py
@@ -1,8 +1,10 @@
 from __future__ import annotations
 
+import pytest
 from inline_snapshot import snapshot
 from kosong.chat_provider.echo import EchoChatProvider
 from kosong.chat_provider.kimi import Kimi
+from kosong.contrib.chat_provider.openai_legacy import OpenAILegacy
 from pydantic import SecretStr
 
 from kimi_cli.config import LLMModel, LLMProvider
@@ -93,3 +95,43 @@ def test_create_llm_requires_base_url_for_kimi():
     model = LLMModel(provider="kimi", model="kimi-base", max_context_size=4096)
 
     assert create_llm(provider, model) is None
+
+
+def test_create_llm_openai_legacy_passes_reasoning_key():
+    provider = LLMProvider(
+        type="openai_legacy",
+        base_url="http://localhost:8000/v1",
+        api_key=SecretStr("dummy"),
+        reasoning_key="reasoning_content",
+    )
+    model = LLMModel(provider="vllm", model="some-model", max_context_size=8192)
+
+    llm = create_llm(provider, model)
+    assert llm is not None
+    assert isinstance(llm.chat_provider, OpenAILegacy)
+    assert llm.chat_provider._reasoning_key == "reasoning_content"
+
+
+def test_create_llm_openai_legacy_reasoning_key_default_none():
+    provider = LLMProvider(
+        type="openai_legacy",
+        base_url="http://localhost:8000/v1",
+        api_key=SecretStr("dummy"),
+    )
+    model = LLMModel(provider="vllm", model="some-model", max_context_size=8192)
+
+    llm = create_llm(provider, model)
+    assert llm is not None
+    assert isinstance(llm.chat_provider, OpenAILegacy)
+    assert llm.chat_provider._reasoning_key is None
+
+
+@pytest.mark.parametrize("value", ["", "  ", "\t\n"])
+def test_reasoning_key_blank_normalized_to_none(value):
+    provider = LLMProvider(
+        type="openai_legacy",
+        base_url="http://localhost:8000/v1",
+        api_key=SecretStr("dummy"),
+        reasoning_key=value,
+    )
+    assert provider.reasoning_key is None