robotnurse.io

./tools/model-fit — a free tool

Which real model can your machine actually run?

Pick a machine — or type the memory straight in — and see which real open-weight models fit, at their actual published GGUF file sizes. No guessing between "7B or 70B": this names the models and the exact quantizations.

Not sure renting beats owning? → cost calculator: API rent vs. your own hardware over 3 years

✎ Accuracy note. The file sizes below are the published GGUF sizes from bartowski's Hugging Face uploads (each model name links to the file listing they came from), fetched 12 September 2026. Quantized file sizes vary between releases — a different llama.cpp version, a different uploader, or a re-quant produces a slightly different file — and your OS, running apps, and context/KV cache all eat into the headroom on top of the file itself. Treat this as a planning ceiling, not a purchase order: before you buy hardware, open the actual file listing for the exact model and quantization you want and read its real size.

What this means for a business

A model that can't fit in memory doesn't just "run a little slower" — it silently degrades or fails outright. When a runtime can't hold the weights, it either refuses to load at all (a hard stop) or spills layers out of memory and crawls to a fraction of its speed — and a non-technical operator reads that crawl as "the AI is slow," not "the machine is too small." For a business, that's the difference between a workflow that holds up all week and a demo that fell apart the first time it met a real queue. Memory isn't a preference you tune later; it's the ceiling on which models you can run at all. Pick the model first, then buy the machine that fits it with headroom — not the other way around.

WANT THIS SET UP FOR YOU?on-site /packages

The matrix tells you what fits. If you'd rather have it running than figured out — the right model, the right quant, the right machine — that's the service.

see the packages →   get the setup audit   or book a call
KEEP THE MATRIX HANDYfree · no spam

New models and quants land constantly. Get the short digest. We'll use your email to deliver what you asked for, plus occasional RobotNurse updates — one-click unsubscribe in every email. Privacy.