Llama
Still the default choice when you need weights you can hold, fine-tune and run offline, with an ecosystem of quantisations and inference stacks built around it. The licence is not OSI-open, and Meta's web presence keeps moving: llama.com now forwards wholesale to the developer site, where Llama is listed next to the newer Muse family, so links rot faster than the models do and the roadmap looks less Llama-first than it did.
Pros
- Weights are downloadable, not just API-accessible
- Broad tooling support: llama.cpp, vLLM, Hugging Face and most inference runtimes
- Multiple sizes so you can trade quality against a single GPU
- Commercial use permitted for almost everyone under the MAU threshold
Cons
- Community licence is not open source; attribution, the Llama name prefix and the MAU cap all bind
- Meta keeps moving the site: llama.com and llama.meta.com now forward to dev.meta.ai, where Llama shares the menu with the Muse models
- Running it is your cost and your operational problem
- Meta's Model API shows no public price; it is login or contact sales