Which fields does the Batch API support when using the OpenAI library to do batch inference via Gemini-3?

When not using Batch API, the ‘extra_body‘ field looks well.

But when using Batch API, it says ‘BadRequestError: Error code: 400 - {‘error’: {‘code’: 400, ‘message’: “Error on line 1: invalid JSON, near column 910: no such field: ‘extra_body’”, ‘status’: ‘INVALID_ARGUMENT’}}‘

Here is a sample of my jsonl:

{"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "gemini-3-pro-preview", "messages": [{"role": "user", "content": "How are you?"}], "temperature": 0.0}, "extra_body": {"extra_body": {"google": {"thinking_config": {"thinking_level": "low", "include_thoughts": true, "reasoning_effort": "low"}}}}}
Any ideas about the issue?