GLM 5.3 runs like a charm in ~240G of VRAM (with half a million tokens context). Depending on what you've got that's between 3 and 10 GPUs. Fewer, larger GPUs will likely run it faster.