Qwen3-Next: Towards Ultimate Training and Inference Efficiency (qwen.ai)

This seems very promising, they are matching Qwen3-235B-A22B-Instruct-2507. It should run on 8-12GB vram with 64 gb ram decently well.