Qwen3.6-35B-A3B strains RTX 4060 rigs

Hot take: the model is efficient for its class, but the hardware ask is still real. The “3B active” headline helps, yet the 35B weight footprint and long-context design mean a single 8GB GPU is more of a compromise box than an ideal deployment target.

• –Official docs show Qwen3.6 targeting hosted APIs and multi-GPU serving paths, with 8-GPU tensor-parallel examples and a 262,144-token default context.
• –An RTX 4060 can likely run a quantized build with CPU offload, but speed and context length will be the first things to collapse.
• –The 32GB system RAM is the saving grace here; it gives you room for offload and larger KV cache, but it does not replace VRAM.
• –For daily local use, a smaller Qwen variant will feel much better; Qwen3.6-35B-A3B is more compelling if you care about capability per parameter than raw responsiveness.
• –Community reaction is already framing it as a serious open release, but also as a model that rewards better hardware and serving frameworks like vLLM, SGLang, or KTransformers.