Qwen 3.6-35B-A3B on dual 5060 Ti with –cpu-moe: 21.7 tok/s at 90K context, with benchmarks vs dense 3.5 and Coder variant
Qwen 3.6 dropped yesterday and I wanted to see if hybrid offloading actually earns its keep on this hardware. My box is two RTX 5060 Ti (32GB VRAM total) with 64GB system RAM. Not a workstation card in sight. I ran the same bench harness across three c…