,
Using the OpenAI ChatGPT robot skill in ARC, how do I invoke the "reasoning" parameter to speed up response time? I want to set GPT-5_Nano's reasoning to minimal. Am I expected to set this in the skill's settings or via ARC scripting (JavaScript/Python)?
Sample code might look like:
reasoning = {"effort": "minimal"}

In ARC’s OpenAI ChatGPT Robot Skill, you don’t set the OpenAI
reasoningparameter through ARC scripting. That skill is designed to use its Settings and Personality tabs, but it does not expose a custom raw JSON request body like:What that means
Recommended approach
If the skill’s UI doesn’t already include a reasoning effort or similar option, then ARC can’t pass it through directly. In that case, the proven path is to:
Important note
GPT-5_nanoand thereasoning.effortparameter are model/API-specific settings. ARC’s OpenAI skill may not currently support passing arbitrary OpenAI request fields unless Synthiam adds that option to the skill.Best way forward
If you need full control over OpenAI request JSON, the better solution is to create a custom robot skill that calls the OpenAI API directly from JavaScript or Python. That lets you send whatever payload the API supports.
Synthiam’s guide for that is here:
https://synthiam.com/Support/Create-Robot-Skill/Overview
Example direction for a custom skill
In a custom skill, you could build the request body yourself and POST it with
Net.hTTPPost().If you want, I can help you with a JavaScript custom skill example that sends
reasoning: { effort: "minimal" }to OpenAI.Also, what are you trying to optimize overall - fast conversational replies, cheaper API usage, or best answer quality with less latency?
Thanks for the info. I am trying to achieve a response time equivalent to the recently decommissioned gpt-4-nano. Gpt-5-nano is far too slow to be of any use. For example, 7 secs to describe an image, whereas gpt-4-nano was 1 sec. As for having a conversation, response time is 30 secs. using gpt-5. By default gpt-5-nano sets the reasoning effort to medium, hence the high latency.
We will take a look to see what parameters are exposed. Be aware that not all parameters are supported by all models. And sometimes the only way to find out is by trying and receiving errors or unusable responses.
Thanks for looking into this. It is an example of ChatGPT OpenAI developers thinking they are improving a service whereas in fact making it worse. If they had kept gpt-4, I would be happy.
This update should help you with your needs
Tested and working. Using the LOW setting, response times were back to gpt-4 response times - about 1 sec for image description. Also I think the responses are more natural than with the MEDIUM setting. Thanks again.