Ruggi.SiteCan my PC run this AI model?
Pick your graphics card or Mac, and see which local AI models fit, which quality level to download, and roughly how fast they will answer. Nothing leaves this page.
Everything runs in this page; nothing is sent anywhere. Estimates are rough (about ±30%): real numbers depend on the program, drivers and settings.
All models
| Model | Size | Best quality that fits | Memory needed | Where it runs | Speed (est.) |
|---|
Your own model
New model not in the list? Enter its numbers from its model page; it appears at the top of the table.
How it works
A model's weights take roughly parameters × bits per weight ÷ 8 of memory (a 4-bit "Q4_K_M" file uses about 4.85 bits per weight). The conversation memory (KV cache) grows with the context length, and the program needs a little extra. If everything fits in the GPU's memory the model runs fast; if it only fits with the help of system RAM it still runs, but slower, because the part in RAM is limited by RAM speed. Mixture-of-experts (MoE) models are large but only use a few "active" parameters per word, so they run faster than their size suggests while still needing the memory to hold all of them.
Data last checked . Parameter counts and layer shapes come from public model pages and are rounded.