Llama

Llama is the most consequential model release of the open-AI era: when Meta published Llama 2's weights under a usable license in 2023, “download a frontier-class model” went from research privilege to Tuesday afternoon. Every Llama generation since — 3, 3.1, 3.2, 3.3, 4 — has pushed the open/closed quality gap narrower, added multimodality, and grown the size ladder so the same family runs on a phone (1B) and a data center (400B).

Its role in this directory is foundational: a huge share of the tools listed here — chat apps, coding agents, local runners like Ollama, serving engines like vLLM, hubs like Hugging Face and ModelScope — exist or thrive because Llama weights are downloadable, fine-tunable and self-hostable. For anyone building on open models, Llama is the default starting point and the benchmark every alternative is measured against.