github/OpenHands
1
1
Fork 0
mirror of https://github.com/All-Hands-AI/OpenHands.git synced 2026-01-10 07:18:10 -05:00
Code Issues Packages Projects Releases Wiki Activity

fix(anthropic): disable extended thinking for Opus 4.1 (#10532)

Browse Source 
Co-authored-by: openhands <openhands@all-hands.dev>
This commit is contained in:
Engel Nyst
2025-08-21 00:13:15 +02:00
committed by GitHub
parent 5ce5469bfa
commit 7861c1ddf7
4 changed files with 33 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
tests/unit/llm/test_llm.py
View File

@@ -1204,7 +1204,7 @@ def test_gemini_medium_reasoning_effort_passes_through(mock_completion):
@patch('openhands.llm.llm.litellm_completion')
def test_opus_41_reasoning_pops_temperature_top_p(mock_completion):
def test_opus_41_keeps_temperature_top_p(mock_completion):
mock_completion.return_value = {
'choices': [{'message': {'content': 'ok'}}],
}
@@ -1217,7 +1217,8 @@ def test_opus_41_reasoning_pops_temperature_top_p(mock_completion):
llm = LLM(config, service_id='svc')
llm.completion(messages=[{'role': 'user', 'content': 'hi'}])
call_kwargs = mock_completion.call_args[1]
assert 'temperature' not in call_kwargs
assert call_kwargs.get('temperature') == 0.7
# Anthropic rejects both temperature and top_p together on Opus; we keep temperature and drop top_p
assert 'top_p' not in call_kwargs
@@ -1239,6 +1240,21 @@ def test_opus_4_keeps_temperature_top_p(mock_completion):
assert call_kwargs.get('top_p') == 0.9
@patch('openhands.llm.llm.litellm_completion')
def test_opus_41_disables_thinking(mock_completion):
mock_completion.return_value = {
'choices': [{'message': {'content': 'ok'}}],
}
config = LLMConfig(
model='anthropic/claude-opus-4-1-20250805',
api_key='k',
)
llm = LLM(config, service_id='svc')
llm.completion(messages=[{'role': 'user', 'content': 'hi'}])
call_kwargs = mock_completion.call_args[1]
assert call_kwargs.get('thinking') == {'type': 'disabled'}
@patch('openhands.llm.llm.litellm.get_model_info')
def test_is_caching_prompt_active_anthropic_prefixed(mock_get_model_info):
# Avoid external calls, but behavior shouldn't depend on model info

4
tests/unit/llm/test_model_features.py
View File

@@ -199,7 +199,6 @@ def test_function_calling_models(model):
'gemini-2.5-flash',
'gemini-2.5-pro',
'gpt-5',
'claude-opus-4-1-20250805',
],
)
def test_reasoning_effort_models(model):
@@ -230,7 +229,6 @@ def test_deepseek_reasoning_effort_models(model):
'claude-3-haiku-20240307',
'claude-3-opus-20240229',
'claude-sonnet-4-latest',
'claude-opus-4-1-20250805',
],
)
def test_prompt_cache_models(model):
@@ -254,7 +252,7 @@ def test_prompt_cache_models(model):
('gemini-2.5-pro', True),
('gpt-5', True),
('gpt-5-2025-08-07', True),
('claude-opus-4-1-20250805', True),
('claude-opus-4-1-20250805', False),
# DeepSeek
('deepseek/DeepSeek-R1-0528:671b-Q4_K_XL', True),
('DeepSeek-R1-0528', True),