Every model needs a specific instruct format and custom sampling settings to work properly, so it's not possible to simply drop in a new one.
I test all major new model releases with this game, and so far only GLM-4-32B-0414 and Gemma 3 27B outperform Nemo, which is the 10 gb/server model. Gemma 3 actively skirts around mature content, so it's not suitable. Both of these models require a GPU with 24 gb of vram. There is no better model than Nemo for 12/16 gb GPUs.
If there's interest in a 24 gb model I will add support for GLM-4.