From dd8df73abd7a419e53d2611f6cee479640bd82a4 Mon Sep 17 00:00:00 2001
From: Michael Stolarz <mstolarz@blaxel.ai>
Date: Wed, 13 May 2026 16:50:01 -0700
Subject: [PATCH] fix livekit reasoning effort default

---
 src/blaxel/livekit/model.py             | 17 ++++++++++
 tests/integration/livekit/test_model.py | 42 +++++++++++++++++++++++++
 2 files changed, 59 insertions(+)

diff --git a/src/blaxel/livekit/model.py b/src/blaxel/livekit/model.py
index 2aea0b42..9848d7c0 100644
--- a/src/blaxel/livekit/model.py
+++ b/src/blaxel/livekit/model.py
@@ -9,6 +9,21 @@
 
 logger = getLogger(__name__)
 
+GPT_54_REASONING_MODELS = {"gpt-5.4", "gpt-5.4-mini", "gpt-5.4-nano"}
+
+
+def _provider_type_value(provider_type):
+    return getattr(provider_type, "value", provider_type)
+
+
+def _set_default_reasoning_effort(provider_type, model: str, kwargs: dict):
+    if (
+        _provider_type_value(provider_type) == "openai"
+        and model in GPT_54_REASONING_MODELS
+        and "reasoning_effort" not in kwargs
+    ):
+        kwargs["reasoning_effort"] = "none"
+
 
 class DynamicHeadersHTTPClient(httpx.AsyncClient):
     """Custom HTTP client that dynamically updates headers on each request."""
@@ -31,6 +46,8 @@ async def send(self, request, *args, **kwargs):
 
 
 async def get_livekit_model(url: str, type: str, model: str, **kwargs):
+    _set_default_reasoning_effort(type, model, kwargs)
+
     # Create custom HTTP client with dynamic headers
     http_client = AsyncOpenAI(
         base_url=f"{url}/v1",
diff --git a/tests/integration/livekit/test_model.py b/tests/integration/livekit/test_model.py
index 9de017c3..c59c2917 100644
--- a/tests/integration/livekit/test_model.py
+++ b/tests/integration/livekit/test_model.py
@@ -1,6 +1,8 @@
 """LiveKit Model Integration Tests."""
 
 pytest_plugins = []
+from unittest.mock import Mock, patch
+
 import pytest  # noqa: E402
 
 pytest.importorskip(
@@ -10,12 +12,52 @@
 from livekit.agents.llm import ChatContext  # noqa: E402
 
 from blaxel.livekit import bl_model  # noqa: E402
+from blaxel.livekit.model import get_livekit_model  # noqa: E402
 
 TEST_MODELS = [
     "sandbox-openai",
 ]
 
 
+async def _created_livekit_kwargs(provider_type: str, model: str, **kwargs):
+    llm = Mock(name="llm")
+    http_client = Mock(name="http_client")
+    openai_client = Mock(name="openai_client")
+
+    with (
+        patch("blaxel.livekit.model.DynamicHeadersHTTPClient", return_value=http_client),
+        patch("blaxel.livekit.model.AsyncOpenAI", return_value=openai_client),
+        patch("blaxel.livekit.model.openai.LLM", return_value=llm) as llm_factory,
+    ):
+        result = await get_livekit_model("https://example.com/models/test", provider_type, model, **kwargs)
+
+    assert result is llm
+    return llm_factory.call_args.kwargs
+
+
+@pytest.mark.asyncio
+async def test_gpt54_openai_models_default_reasoning_effort_to_none():
+    for model_name in ["gpt-5.4", "gpt-5.4-mini", "gpt-5.4-nano"]:
+        kwargs = await _created_livekit_kwargs("openai", model_name)
+
+        assert kwargs["model"] == model_name
+        assert kwargs["reasoning_effort"] == "none"
+
+
+@pytest.mark.asyncio
+async def test_livekit_reasoning_effort_override_is_preserved():
+    kwargs = await _created_livekit_kwargs("openai", "gpt-5.4-mini", reasoning_effort="low")
+
+    assert kwargs["reasoning_effort"] == "low"
+
+
+@pytest.mark.asyncio
+async def test_livekit_non_openai_provider_does_not_default_reasoning_effort():
+    kwargs = await _created_livekit_kwargs("xai", "gpt-5.4-mini")
+
+    assert "reasoning_effort" not in kwargs
+
+
 @pytest.mark.asyncio(loop_scope="class")
 class TestBlModel:
     """Test bl_model functionality."""