diff --git a/deeptutor/services/llm/factory.py b/deeptutor/services/llm/factory.py
index 4fe8a97d1..6345f5cf5 100644
--- a/deeptutor/services/llm/factory.py
+++ b/deeptutor/services/llm/factory.py
@@ -272,6 +272,8 @@ def _sanitize_call_kwargs(
 
     if not supports_response_format(binding, model):
         extra_kwargs.pop("response_format", None)
+    if binding == "azure_openai" and "max_completion_tokens" in extra_kwargs:
+        extra_kwargs["max_tokens"] = extra_kwargs.pop("max_completion_tokens")
     return extra_kwargs
 
 
diff --git a/tests/services/llm/test_factory_provider_exec.py b/tests/services/llm/test_factory_provider_exec.py
index 8f2aa95d8..a6ba1c5c5 100644
--- a/tests/services/llm/test_factory_provider_exec.py
+++ b/tests/services/llm/test_factory_provider_exec.py
@@ -201,6 +201,28 @@ async def test_complete_strips_unsupported_response_format(monkeypatch) -> None:
     assert "response_format" not in provider.complete_kwargs
 
 
+@pytest.mark.asyncio
+async def test_complete_normalizes_azure_max_completion_tokens(monkeypatch) -> None:
+    cfg = _make_cfg(
+        model="gpt-5.4",
+        binding="azure_openai",
+        provider_name="azure_openai",
+    )
+    provider = _FakeProvider()
+
+    monkeypatch.setattr("deeptutor.services.llm.factory.get_llm_config", lambda: cfg)
+    monkeypatch.setattr(
+        "deeptutor.services.llm.factory.get_runtime_provider",
+        lambda _config: provider,
+    )
+
+    result = await complete("hello", max_completion_tokens=200)
+
+    assert result == "ok"
+    assert provider.complete_kwargs["max_tokens"] == 200
+    assert "max_completion_tokens" not in provider.complete_kwargs
+
+
 @pytest.mark.asyncio
 async def test_complete_passes_retry_delays(monkeypatch) -> None:
     cfg = _make_cfg()