The discussion revolves around the support of GPT-4o mini in Discourse, with users curious about its implementation and cost-effectiveness compared to GPT-3.5 Turbo 1. One user confirms it’s already supported via the Discourse AI plugin by configuring gpt-4o-mini in the LLM settings 3. Another user expresses embarrassment for not knowing this earlier 4. Some users discuss potential hallucination issues with newer models 5, while others question the separation of model configuration between general settings and the LLM section 5. A contributor explains the current setup allows users to configure LLMs separately before enabling them for specific AI features like the AI bot, noting internal discussions about streamlining this process 7.
Title says it all – just curious if there will be a plan to support this model and if so, is there a rough estimate on when it could be implemented?
This is a massive improvement for anyone currently using GPT 3.5 Turbo as it’s smarter and over 60% cheaper.
It already works with the Discourse AI plugin, you just need to make sure to use gpt-4o-mini as the name in your LLM configuration:
Well this is embarrassing. Thanks boss 
Perhaps I haven’t played around with it enough - note made
We will see how much smarter it is or much better it is to hallucinate 
I remember how much buzz there was when GPT-4o came and now out there is lot of complains.
But good to try.
Bit off topic, but I don’t totally understand why we have two places to tell model and API, settings and LLM-section?
That’s a good point. I suspect the hallucination problems couldn’t have improved with this model however I managed to largely mitigate that by putting a bunch of constraints throughout the system prompt – though that has its own downsides of course.
It makes sense only for the manual LLM set up. So I find myself asking the same question 
The idea is that you setup your LLM 1st and then pick that LLM for the Discourse AI feature of your choosing hence why they are currently in 2 separate places
We have had some internal discussions regarding perhaps having this process all in one place i.e - right when you setup the LLM you can also toggle it on for the specific AI features, like how we have it for AI bot
