Which open-weight LLMs can your hardware run?

VRAM estimates are computed as parameters × bytes-per-quant × 1.2 overhead, picking the best available quantization for your budget. Apple Silicon figures apply a 0.75 usable-memory factor since the OS reserves headroom in unified memory. CPU-only inference is supported but expect slower generation. Methodology →

Enter your VRAM

Pick a preset rig or type a custom VRAM budget to see which models fit — ranked by intelligence, speed, and price.

CPU-only inference is correct but slow. Expect 1–5 tokens/sec on a modern desktop CPU.