run a model

Run a model at home

Two ways in, same arithmetic. Name the model you want and we work out the hardware it takes; or name the hardware you already have and we work out what fits in its memory. Plus what generation adds to your electricity bill, which nobody else counts.

models people run

Pick the one you want to run, and we work out the hardware it takes — from the smallest that manages it to the one that runs it without thinking.

Qwen3.5 4B2.7 GB needed Llama 3.1 8B5.4 GB needed Qwen3.5 9B6.1 GB needed Gemma 4 12B8.2 GB needed Phi-4 14B9.5 GB needed gpt-oss 20B13.6 GB needed Gemma 4 26B-A4B17.1 GB needed Qwen3.6 27B18.4 GB needed Qwen3-Coder 30B-A3B20.4 GB needed Gemma 4 31B20.9 GB needed DeepSeek-R1 Distill 32B21.8 GB needed Qwen3.6 35B-A3B23.8 GB needed Llama 3.1 70B47.6 GB needed gpt-oss 120B81.6 GB needed

or the other way round: what your hardware can run

Already have a card, a Mac, or nothing but a processor? Start here instead — same arithmetic, read from the other end.

graphics cards

apple silicon

Apple Silicon, 16 GB unified12 GB usable Apple Silicon, 24 GB unified18 GB usable Apple Silicon, 32 GB unified24 GB usable Apple Silicon, 48 GB unified36 GB usable Apple Silicon, 64 GB unified48 GB usable Apple Silicon, 128 GB unified96 GB usable

no graphics card

No graphics card, 16 GB of RAM9.6 GB usable No graphics card, 32 GB of RAM19.2 GB usable No graphics card, 64 GB of RAM38.4 GB usable

builder

Price a whole machine around itA card is not a server. The builder adds the machine, the drives, the electricity and the backup.