I was curious what everyone’s preferred LLMs to use for your Agents and how you decide which one to use.
When I’m building or running sessions in Brim, I usually just stick with the default models that Brim suggest for the Agents. Apart from writing which I tend to switch to Claude because I just like how it writes.
I’m wondering how others approach this. Do you mostly leave the default on, or do you switch models depending on the task? And if you do switch, how do you decide which one is best for what you’re doing?
Would be interesting to hear how others are thinking about this in practice.
In Brim, we usually recommend sticking with the default models the platform suggests. Those are chosen deliberately based on the task, our testing, and what tends to deliver the most reliable outcome in practice.
As a rough rule of thumb:
Use GPT style models for reasoning, and decision making
Use Gemini style models when speed matter
Use Claude style models when quality of writing and tone really matter
That said, we actively test different models internally using evals for specific tasks, and the best model often depends on the exact job you are asking the agent to do.
If you are ever unsure or want to go deeper, feel free to DM me with the task you are trying to accomplish and we are happy to share how we would approach model selection or what the evals show.