Running a local LLM is entirely about VRAM: it decides which model sizes fit, and nothing else comes close. A mini-PC only helps here if it has a real PCIe slot for a GPU - the iGPU will not do useful inference - and most mini-PCs simply do not, which narrows the field fast.

So this list is short by design: only boxes that can physically take a fitting low-profile GPU appear, ranked by the VRAM that GPU brings. If you want a big card, you are into a DIY build with a real case and PSU rather than a mini-PC.

What to look for

A real PCIe slot
x8 or x16 that takes a low-profile card. Without it, a mini-PC is a non-starter for GPU inference.
VRAM sizing
~8GB runs a 7B model, ~12GB a 13B, ~24GB a 33B (all Q4). Buy for the model you actually want to run.
Slot power ~75W
A mini-PC slot supplies about 75W with no external connector, so low-profile cards like the A2000 are the natural fit.
Or go DIY
Want a 16GB+ full-size GPU? That is a proper build, not a mini-PC - budget for a case and PSU.

The picks right now

# Pick Fit Price
1 Lenovo ThinkCentre M720q Tiny + Nvidia RTX PRO 4000 Blackwell SFF Top pick
Nvidia RTX PRO 4000 Blackwell SFF fits here - up to ~33B (Q4).
24 GB VRAM
~$1,620 Check price on Amazon
2 Lenovo ThinkCentre M920q Tiny + Nvidia RTX PRO 4000 Blackwell SFF
Nvidia RTX PRO 4000 Blackwell SFF fits here - up to ~33B (Q4).
24 GB VRAM
~$1,650 Check price on Amazon
3 Lenovo ThinkCentre M70q Gen 1 + Nvidia RTX PRO 4000 Blackwell SFF
Nvidia RTX PRO 4000 Blackwell SFF fits here - up to ~33B (Q4).
24 GB VRAM
~$1,675 Check price on Amazon

Ranked from curated specs, same logic as the interactive recommender. Approximate - verify against your exact setup. Prices move; check the retailer.

Other guides