20GB VRAM + 64GB DDR5 - Qwen3.6 35B A3B still the best choice?

Hey everyone. Trying to figure out the best setup for my hardware. I've got a 64GB DDR5 RAM laptop and A 20GB 7900xt egpu. Usecase is only pi-coding-agent locally. So far, Qwen3.6 35B A3B has been the only model I've found that would work well with CPU offload (MoE) at higher quants - I've managed to fit Q8 at 100k context or ud-q6kl at full 262k context. It works okay, on the slower side but for me anything with 20+tks is very useable. However, I'm interested in what you guys are running on similar hardwar