`ollama run` and you have a capable model on your machine, offline, with no per-token cost. If you run anything at volume — tagging, drafting, summarising — this is the difference between a hobby and a margin.
Model quality scales with the RAM you have. On 8 GB you get toys; the useful ones want 16–32 GB+.
| What you get free | Free and open source (MIT); models download once and run offline forever |
| Where the limit sits | No limits beyond your own hardware — bigger models want serious RAM |
| Credit card to start | No |
| Pricing type | Open source |
| Last checked | 2026-07-30 · source |
| Run it yourself | Yes — MIT · repo |
| Platforms | windows, mac, linux, api |
| Category | Self-hosted / local · Code & dev · Chat & assistants |
We recheck every entry. Subscribers hear first when a free tier quietly shrinks.