From fede37b49620c63d536ea8197717b87a14fe4fa9 Mon Sep 17 00:00:00 2001
From: Clay Arnold <99657852+314clay@users.noreply.github.com>
Date: Thu, 19 Feb 2026 12:33:17 +1300
Subject: [PATCH] fix: add claude-opus-4-6 to temperature/top_p guard (#12874)

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
---
 openhands/llm/llm.py       |  5 +++--
 tests/unit/llm/test_llm.py | 19 +++++++++++++++++++
 2 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/openhands/llm/llm.py b/openhands/llm/llm.py
index cbbac29c1e..1a0734a6ee 100644
--- a/openhands/llm/llm.py
+++ b/openhands/llm/llm.py
@@ -198,13 +198,14 @@ class LLM(RetryMixin, DebugMixin):
         if 'claude-opus-4-1' in self.config.model.lower():
             kwargs['thinking'] = {'type': 'disabled'}
 
-        # Anthropic constraint: Opus 4.1, Opus 4.5, and Sonnet 4 models cannot accept both temperature and top_p
+        # Anthropic constraint: Opus 4.1, Opus 4.5, Opus 4.6, and Sonnet 4 models cannot accept both temperature and top_p
         # Prefer temperature (drop top_p) if both are specified.
         _model_lower = self.config.model.lower()
-        # Apply to Opus 4.1, Opus 4.5, and Sonnet 4 models to avoid API errors
+        # Apply to Opus 4.1, Opus 4.5, Opus 4.6, and Sonnet 4 models to avoid API errors
         if (
             ('claude-opus-4-1' in _model_lower)
             or ('claude-opus-4-5' in _model_lower)
+            or ('claude-opus-4-6' in _model_lower)
             or ('claude-sonnet-4' in _model_lower)
         ) and ('temperature' in kwargs and 'top_p' in kwargs):
             kwargs.pop('top_p', None)
diff --git a/tests/unit/llm/test_llm.py b/tests/unit/llm/test_llm.py
index b04425e631..e27bc354d9 100644
--- a/tests/unit/llm/test_llm.py
+++ b/tests/unit/llm/test_llm.py
@@ -1274,6 +1274,25 @@ def test_opus_45_keeps_temperature_drops_top_p(mock_completion):
     assert 'top_p' not in call_kwargs
 
 
+@patch('openhands.llm.llm.litellm_completion')
+def test_opus_46_keeps_temperature_drops_top_p(mock_completion):
+    mock_completion.return_value = {
+        'choices': [{'message': {'content': 'ok'}}],
+    }
+    config = LLMConfig(
+        model='anthropic/claude-opus-4-6',
+        api_key='k',
+        temperature=0.7,
+        top_p=0.9,
+    )
+    llm = LLM(config, service_id='svc')
+    llm.completion(messages=[{'role': 'user', 'content': 'hi'}])
+    call_kwargs = mock_completion.call_args[1]
+    assert call_kwargs.get('temperature') == 0.7
+    # Anthropic rejects both temperature and top_p together on Opus 4.6; we keep temperature and drop top_p
+    assert 'top_p' not in call_kwargs
+
+
 @patch('openhands.llm.llm.litellm_completion')
 def test_sonnet_4_keeps_temperature_drops_top_p(mock_completion):
     mock_completion.return_value = {