Hi, I am using

openai/gpt-oss-120b

on Groq provider, and I getting quite poor results compared to other similar models like gpt 4.1 (have similar results on benchmarks). I suspect that its because it has reasoning turned off completely because in the logs I dont see any reasoning tokens. Is it possible to turn on reasoning and control Reasoning Effort (because I dont see it anywhere in the docs)?

Hi @N3squik, The groq provider in Vapi only provides the following options for configuration: https://docs.vapi.ai/api-reference/assistants/create#request.body.model.Groq, and unfortunately, enabling or disabling reasoning is not an configurable option. A good workaround for your usecase would be to use Custom LLM where your server will act like a proxy between Vapi and groq. This will enable you to configure your LLM as per your need.

Thank you for the advice. Do you know how Vapi is calling Groq's

openai/gpt-oss-120b

, can you confirm if reasoning is turned off?

Hi @N3squik, I have checked it from my side and yes, Vapi sends the request with reasoning turned off. Might be due to how the default settings are handled during the API calls.