In Ollama, models supported by Llama.cpp are basically usable, so you can use almost any architecture model in the same way.
One thing to note about recent models is that Qwen 3 seems to perform worse with quantization smaller than Q6, so be careful.