GPT-5-Nano accepted parameters

API
1

Does GPT-5-Nano not support the reasoning param?

I just got a 400 for sending reasoning: { effort: “minimal” } saying that reasoning is an unknown parameter.

I guess my question is, do we have any say in nano or mini’s reasoning or is that just for gpt-5?

2

Yes, gpt-5-nano is a reasoning model, and on both Chat Completions and on the Responses endpoint, it supports that endpoint’s particular version of sending the parameter to limit or expand the reasoning.

Responses, a snippet I have open just now:

if reasoning_model:
    # Reasoning models forbid sampling params; send reasoning object instead
    payload["reasoning"] = {
        "effort": "minimal",  # one of: "minimal", "low", "medium", "high"
        "summary": None,  # "detailed" except on codex/cu models
    }
else:
    # Normal models accept temperature / top-p for quality control
    payload["temperature"] = 0.5
    payload["top_p"] = 0.9

Chat Completions, its own API parameter:
"reasoning_effort":"minimal",

Your reasoning billing after parameter and a bit more developer prompting:

= RESTART: responses-simple-gpt-5-conversation.py
--Usage--  in/cached: 66/0;  out/reasoning:69/0

Assistant: Hey! I’m Jade—a witty, sarcastic Gen Alpha teen. I’m powered by a snazzy AI brain (think GPT-5-nano vibes) and I’m here to dish quick takes, witty comebacks, and help with whatever you need—short and sweet. What’s up?

Prompt (or 'exit'): how long do you think before a final response to me?
--Usage--  in/cached: 150/0;  out/reasoning:70/0

Assistant: Honestly? usually just a few seconds. I’ll aim for near-instant, but if I’m juggling a big question or some tricky info, I’ll pause to make sure it’s solid. Either way, I’ll tell you if I need a moment. Want me to time my replies to keep it snappy?

OpenAI SDKs will block parameters they don’t know – to avoid an upgrade treadmill, block openai libraries from your workflow.