Title: reasoning_effort: "minimal" not working for Gemini 3 Flash despite documentation

Gemini API
, , ,
1

I’m trying to use the OpenAI compatibility layer with Gemini 3 Flash Preview and the reasoning_effort parameter. According to the official documentation, minimal should be a valid value that maps to thinking_level=minimal for Gemini 3 Flash:

image
image2214×952 185 KB

However, when I try to use reasoning_effort=“minimal”, I get:

`BadRequestError: Error code: 400 - [{‘error’: {‘code’: 400, ‘message’: ‘Invalid reasoning_effort: minimal. Valid values are: high, low, medium, none’, ‘status’: ‘INVALID_ARGUMENT’}}]

Minimal reproduction:

import asyncio
from openai import AsyncOpenAI

GOOGLE_AI_API_KEY = "your-api-key"
GOOGLE_AI_OPENAI_BASE = "https://generativelanguage.googleapis.com/v1beta/openai/"


async def test_reasoning_effort(effort: str):
    print(f"\n--- Testing reasoning_effort='{effort}' ---")

    client = AsyncOpenAI(
        api_key=GOOGLE_AI_API_KEY,
        base_url=GOOGLE_AI_OPENAI_BASE,
        max_retries=0,
    )

    try:
        response = await client.chat.completions.create(
            model="gemini-3-flash-preview",
            reasoning_effort=effort,
            messages=[
                {"role": "system", "content": "You are a helpful assistant."},
                {"role": "user", "content": "What is 2+2?"},
            ],
        )
        print(f"SUCCESS: {response.choices[0].message.content[:100]}...")
    except Exception as e:
        print(f"FAILED: {type(e).__name__}: {e}")


async def main():
    for effort in ["minimal", "low", "medium", "high", "none"]:
        await test_reasoning_effort(effort)


if __name__ == "__main__":
    asyncio.run(main())

Result:

--- Testing reasoning_effort='minimal' ---
FAILED: BadRequestError: Invalid reasoning_effort: minimal. Valid values are: high, low, medium, none

--- Testing reasoning_effort='low' ---
SUCCESS: 2 + 2 = 4...

--- Testing reasoning_effort='medium' ---
SUCCESS: 2 + 2 = 4...

--- Testing reasoning_effort='high' ---
SUCCESS: 2 + 2 = 4...

--- Testing reasoning_effort='none' ---
SUCCESS: 2 + 2 = 4...

he API error message explicitly says valid values are high, low, medium, none - but minimal is not in that list despite being documented.

Environment:

  • openai SDK version: 2.15.0 (current)

  • Date: January 22, 2026

Is this a documentation error, or is minimal support not yet deployed? Its quite unfortunate as for our use-case more thinking often correlates to higher errors.

1 Like