https://kunbuilds.com/en/ Notes on local models, agents and developer tools. en https://kunbuilds.com/en/posts/qwen3.6-35b-a3b-6gb-llama-cpp/ https://kunbuilds.com/en/posts/qwen3.6-35b-a3b-6gb-llama-cpp/ With enough system RAM, a 35B-class MoE model can be practically usable with 6 GB of VRAM, provided its expert weights stay on the CPU.