Unable to get Leo AI BYOM to work with OpenAI gpt-5 model

Brave Browser Leo AI
,
1

I am trying to use the new gpt-5 model from OpenAI in the BYOM (bring your own model) mode of Leo AI, but it results in an error message “There was a network issue connecting to Leo, check your connection and try again.“

I have been able to successfully use Leo AI BYOM with gpt-4o and gpt-4.1 with the same API Key, and I’ve verified using curl that gpt-5 works fine with that same API Key. But not within Leo AI. Could it be that Brave automatically passes a temperature value to https://api.openai.com/v1/chat/completions because that does not work with gpt-5 (but works with all the older models)? If so, then that should be fixed or at least the temperature should be made into a parameter that can be adjusted in settings.

2

Hey,

Thanks for letting us know, seems like you could be right - https://github.com/brave/brave-core/blob/c3fffdfb10c274b45eab35495c5ac142b505de5c/components/ai_chat/core/browser/engine/oai_api_client.cc#L67

dict.Set(“temperature”, 0.7);

I’ll raise an issue and try to get it fixed, sorry about this!

3

It seems like you understand the issue, but for others following along, this error occurs because GPT-5 accessed through the OpenAI API does not support the temperature parameter.

By the way, it would also be nice to be able to customize the reasoning_effort parameter when using a model that supports it.

4

If you don’t want to hold your breath waiting for Brave to update Leo to support GPT-5, I found this simple tool – kanna5/brave-byom-proxy – that routes requests to the OpenAI API through a local proxy, and strips the temperature parameter so you can use gpt-5 as the model.

To run it:

  1. Install and run:
git clone https://github.com/kanna5/brave-byom-proxy.git
cd brave-byom-proxy
uv sync
uv run fastapi run main.py
  1. Add a new BYOM model in Leo settings:
  • Model request name: gpt-5.1
  • Server endpoing: http://localhost:8000/v1/chat/completions?reasoning_effort=low
  • Context size: 400000
  • Vision support: enabled