[cat_plus_plus]

That's why I run a local Qwen3-Next model on an NVIDIA Thor dev kit (Apple Silicon and DGX Spark are other options but they are even more expensive for 128GB VRAM)