What can my computer run?
The first question everyone asks about local AI — answered honestly for your exact machine. Four quick questions, then a verdict for every model size: what runs comfortably, what’s a squeeze, and what won’t fit.
No benchmarks, no tokens-per-second charts — plain words. And when we can’t size something (a graphics card we don’t know), we say so instead of guessing.
Tell us about your computer
Four quick questions — no account needed, and it’s saved on this device only. You can change any answer anytime and the list below updates instantly.
Your machine’s tier list
Answer the quick questions above— this list fills in instantly, for every model size at once.
The three terms worth knowing
- Parameters (the “7B” in a model name)
- A model’s size, in billions of internal values. More parameters generally means smarter answers — and more memory needed to hold the model.
- Q4 and Q8 (quantization)
- Compressed versions of the same model. Q4 is the standard download — roughly half the size of Q8 with a small quality trade-off most people never notice. Apps like Ollama and LM Studio hand you Q4 by default, which is why our verdicts size against it.
- VRAM and unified memory
- Where the model actually lives while running. On a PC that’s the graphics card’s own memory (VRAM); on an Apple Silicon Mac the processor and graphics share one pool (unified memory), so a 32 GB Mac can host models a 16 GB graphics card can’t.
Ready to try one? Ollama and LM Studio are the friendliest starting points — both pick a sensibly-sized download for your machine automatically.
