Qwen3.6 27B

User reports Qwen 3.6 27B at ~24 t/s on dual RTX 3060 GPUs. The user also discusses diffusion-based techniques, MOE, quants, and token authority as a wishlist for future local models.