Llama 3.1

Meta released Llama 3.1 in July 2024 as an open weight language model family in 8B, 70B, and 405B parameter sizes, each in pretrained and instruction tuned form. The models support a 128K token context window, multilingual text, and built-in tool calling, and the 405B variant was among the strongest openly available models at launch. Weights are distributed through Meta's site and Hugging Face under the Llama 3.1 Community License, which permits research and commercial use subject to an acceptable use policy and a usage threshold for the very largest deployments. Alongside the release, Meta reorganized its GitHub presence around the Llama Stack, splitting model utilities, safety tooling, and agent components into separate repositories and deprecating the original llama3 repo. Developers and businesses use Llama 3.1 as a base for fine-tuning, distillation to smaller models, and self hosted deployments where data control matters.