User asks about Qwen3.6 for local coding agent on 20GB VRAM

Reddit user with 20GB VRAM (7900xt eGPU) and 64GB DDR5 seeks best model for a pi-coding-agent locally. They find Qwen3.6 35B A3B performs well with CPU offload at higher quants, but ask if alternatives exist.