ai-agents-thinking-parameter-s-when-using-claude-3-7-sonnet-with-the-thinking-par-29c4804e

THEN Use the model name directly (e.g., 'claude-3-7-sonnet-20250219') with the proxy's base URL instead of the 'litellm_proxy/<model>' prefix. Ensure that 'budget_tokens' is at least 1024 and 'temperature' is set to 1 when enabling thinking. Alternatively, set 'litellm.drop_params=True' or for the proxy configure 'litellm_settings: drop_params: true' to silently drop unsupported parameters.

response = litellm.completion(
    api_key="sk-...",
    base_url="http://localhost:4000",
    model='claude-3-7-sonnet-20250219',
    messages=[{"role": "user", "content": "..."}],
    max_tokens=2001,
    temperature=1,
    thinking={"type": "enabled", "budget_tokens": 2000}
)