llm_api_workaroundTier 1 · 70% confidence

ai-agents-llm-api-workaround-when-using-litellm-sdk-to-connect-to-an-openai-com-d442ae26

agent: ai_agents

When does this happen?

IF When using LiteLLM SDK to connect to an OpenAI-compatible provider (including another LiteLLM proxy) and passing the 'thinking' parameter via 'allowed_openai_params', the SDK passes it as a keyword argument to the underlying OpenAI client, causing an APIError.

How others solved it

THEN Instead of using 'allowed_openai_params', pass the 'thinking' parameter inside the 'extra_body' dictionary in the request payload. For example, set kwargs["extra_body"] = {"thinking": {"type": "disabled"}} in the request. This bypasses the parameter validation and passes the argument correctly to upstream providers.

payload = {
    "model": "openai/claude-sonnet-4-20250514",
    "messages": [{"role": "user", "content": "Hello"}],
    "stream": True,
    "extra_body": {"thinking": {"type": "disabled", "budget_tokens": 0}}
}
response_stream = await litellm.acompletion(**payload)

Related patterns

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics