Fast enough depends what you are doing. Models down around 8B params will fit on the card, Ollama can spill out though so if you need more quality and can toler

Fast enough depends what you are doing. Models down around 8B params will fit on the card, Ollama can spill out though so if you need more quality and can tolerate the latency bigger models like the 30B MoE might be good. I don't have much experience with Qwen3 but Qwen2.5 coder 7b and Gemma3 27b are examples of those two paths that I've used a fair amount.