Run your own AI · Will it run on my machine?
Will it run on my machine?
thecontinuitylayer.com/open/will-it-runChecked 22 October 2026. US dollars. No affiliate links.
Memory decides which models fit. Speed decides whether you'll enjoy it: below about 5 to 10 words a second it feels painful. First, check your machine with CanIRun.ai. Then come back here to see what your result means.
What fits, by machine
| Your machine | What fits | Good for |
|---|---|---|
| 8GB laptop or 8GB graphics card | Small 4–8B models | Basic chat, short tasks |
| 16GB laptop or Mac | Up to about 12B; only 9–12GB is really usable | Chat and documents; agents will disappoint |
| 24–32GB Mac | Qwen3.8-27B, Gemma 4 26B | Solid drafts and summaries |
| 24GB graphics card | 27–32B models, about 37–71 words a second measured | Good everyday local AI |
| 128GB box | gpt-oss-120b, about 55–88 words a second measured | Close to a hosted experience |
Before you buy hardware
- Don't buy a graphics card under 12GB for this.
- For a 128GB machine, the Mac is fastest and quietest, Strix Halo is cheapest, and DGX Spark is poor value since its price rose to $6,950 on 2 October 2026.
- If you only want better answers, a $10–20 plan is cheaper than any hardware.
Which model to pick for your tier is in the model finder; what each kind of machine is good at, and when to rent instead, is in the hardware guide.
Speeds come from ComputingForGeeks and its 128GB comparison.