Llama
Meta's family of language models whose weights are downloadable, which made it a foundation for a huge ecosystem of open models, fine-tunes and local AI.
Also LLaMA · open-weight models
Touvron et al., Meta AI, 2023
Most big AI models live only on their company's servers. Meta handed out the model files themselves, so anyone with a strong enough computer could run, study and adapt them.
Where it fits
+All relationships as a list
Is a kind of / part of
Siblings
Uses / built on
How it works
In February 2023, Meta AI published LLaMA: decoder-only Transformers from 7 to 65 billion parameters, trained only on publicly available data. The paper’s headline: the 13B model beat the far larger GPT-3 on most benchmarks, by training a smaller model on more data, as scaling laws research had suggested.
The bigger impact was access. Researchers could download the weights, and the community quickly learned to fine-tune them cheaply with LoRA and run them on laptops with quantization. Later Llama generations were released under licenses allowing broad commercial use (with conditions), and Llama models became a common base for open fine-tunes worldwide.
Sizes, versions and licenses change frequently, so check the current release before relying on specifics.
Everyday use cases
- Case 01
How you've probably met it
Through apps and research projects built on fine-tuned Llama models, or by running one locally with community tools.
Strengths, weaknesses and when to use it
Strengths
- Open weights enable local use, research and fine-tuning without an API.
Weaknesses
- License terms have conditions; "open weights" is not the same as fully open source.
When to use itWhen you need a model you can run, inspect or fine-tune yourself.
Compare with
Go deeper
Sources
Facts checked as of 2026-10-12