Set default temperature to None instead of 0.0

openhands-sdk/openhands/sdk/llm/llm.py

@@ -180,7 +180,9 @@ class LLM(BaseModel, RetryMixin, NonNativeToolCallingMixin):
         ge=0,
         description=(
             "Sampling temperature for response generation. "
-            "Defaults to 0 for most models and provider default for reasoning models."
+            "Defaults to None (uses provider default temperature). "
+            "Set to 0.0 for deterministic outputs, "
+            "or higher values (0.7-1.0) for more creative responses."
         ),
     )
     top_p: float | None = Field(default=1.0, ge=0, le=1)
@@ -461,7 +463,7 @@ def _set_env_side_effects(self):
         self._init_model_info_and_caps()
 
         if self.temperature is None:
-            self.temperature = get_default_temperature(self.model)
+            self.temperature = get_default_temperature()
 
         logger.debug(
             f"LLM ready: model={self.model} base_url={self.base_url} "

openhands-sdk/openhands/sdk/llm/utils/model_features.py

@@ -180,20 +180,11 @@ def get_features(model: str) -> ModelFeatures:
     )
 
 
-# Default temperature mapping.
-# Each entry: (pattern, default_temperature)
-DEFAULT_TEMPERATURE_MODELS: list[tuple[str, float]] = [
-    ("kimi-k2-thinking", 1.0),
-    ("kimi-k2.5", 1.0),
-]
-
-
-def get_default_temperature(model: str) -> float:
-    """Return the default temperature for a given model pattern.
+def get_default_temperature() -> float | None:
+    """Return the default temperature for a given model.
 
-    Uses case-insensitive substring matching via model_matches.
+    Returns None to use provider defaults.
+    Previously supported specific temperature overrides for certain models,
+    but now all models use provider defaults.
     """
-    for pattern, value in DEFAULT_TEMPERATURE_MODELS:
-        if model_matches(model, [pattern]):
-            return value
-    return 0.0
+    return None

tests/sdk/config/test_llm_config.py

@@ -21,7 +21,7 @@ def test_llm_config_defaults():
     assert config.retry_max_wait == 64
     assert config.timeout == 300  # Default timeout is 5 minutes
     assert config.max_message_chars == 30_000
-    assert config.temperature == 0.0
+    assert config.temperature is None  # None to use provider defaults
     assert config.top_p == 1.0
     assert config.top_k is None
     assert config.max_input_tokens == 128000  # Auto-populated from model info

tests/sdk/llm/test_llm.py

@@ -637,7 +637,7 @@ def test_llm_local_detection_based_on_model_name(default_llm):
 
     # Test basic model configuration
     assert llm.model == "gpt-4o"
-    assert llm.temperature == 0.0
+    assert llm.temperature is None  # Uses provider default
 
     # Test with localhost base_url
     local_llm = default_llm.model_copy(update={"base_url": "http://localhost:8000"})

tests/sdk/llm/test_model_features.py

@@ -314,52 +314,17 @@ def test_send_reasoning_content_support(model, expected_send_reasoning):
     assert features.send_reasoning_content is expected_send_reasoning
 
 
-@pytest.mark.parametrize(
-    "model,expected_temperature",
-    [
-        # kimi-k2-thinking models should default to 1.0
-        ("kimi-k2-thinking", 1.0),
-        ("kimi-k2-thinking-0905", 1.0),
-        ("Kimi-K2-Thinking", 1.0),  # Case insensitive
-        ("moonshot/kimi-k2-thinking", 1.0),  # With provider prefix
-        ("litellm_proxy/kimi-k2-thinking", 1.0),  # With litellm proxy prefix
-        # kimi-k2.5 models should also default to 1.0
-        ("kimi-k2.5", 1.0),
-        ("Kimi-K2.5", 1.0),  # Case insensitive
-        # All other models should default to 0.0
-        ("kimi-k2-instruct", 0.0),  # Different kimi variant
-        ("gpt-4", 0.0),
-        ("gpt-4o", 0.0),
-        ("gpt-4o-mini", 0.0),
-        ("claude-3-5-sonnet", 0.0),
-        ("claude-3-7-sonnet", 0.0),
-        ("gemini-1.5-pro", 0.0),
-        ("gemini-2.5-pro-experimental", 0.0),
-        ("o1", 0.0),
-        ("o1-mini", 0.0),
-        ("o3", 0.0),
-        ("deepseek-chat", 0.0),
-        ("llama-3.1-70b", 0.0),
-        ("azure/gpt-4o-mini", 0.0),
-        ("openai/gpt-4o", 0.0),
-        ("anthropic/claude-3-5-sonnet", 0.0),
-        ("unknown-model", 0.0),
-    ],
-)
-def test_get_default_temperature(model, expected_temperature):
-    """Test that get_default_temperature returns correct values for different models."""
-    assert get_default_temperature(model) == expected_temperature
+def test_get_default_temperature():
+    """Test that get_default_temperature returns None."""
+    assert get_default_temperature() is None
 
 
 def test_get_default_temperature_fallback():
-    """Test that get_default_temperature returns 0.0 for unknown models."""
-    assert get_default_temperature("completely-unknown-model-12345") == 0.0
-    assert get_default_temperature("some-random-model") == 0.0
+    """Test that get_default_temperature returns None."""
+    assert get_default_temperature() is None
 
 
-def test_get_default_temperature_case_insensitive():
-    """Test that get_default_temperature is case insensitive."""
-    assert get_default_temperature("kimi-k2-thinking") == 1.0
-    assert get_default_temperature("KIMI-K2-THINKING") == 1.0
-    assert get_default_temperature("Kimi-K2-Thinking") == 1.0
-    assert get_default_temperature("KiMi-k2-ThInKiNg") == 1.0
+def test_get_default_temperature_returns_none():
+    """Test that get_default_temperature returns None regardless of input."""
+    # All models now return None, so case doesn't matter
+    assert get_default_temperature() is None