Meta Unveils Llama 3.1, a Frontier-Level Open Source AI Model
Meta releases Llama 3.1, featuring a massive 405B model that rivals top closed-source systems. The updated suite brings longer context windows, multilingual support, and new tools for building custom AI agents.
Meta introduces Llama 3.1, marking a major milestone for openly accessible artificial intelligence. The flagship model, Llama 3.1 405B, stands as the world's largest and most capable open source foundation model, offering state-of-the-art performance in general knowledge, math, tool use, and multilingual translation that directly rivals top closed-source alternatives.
The updated model suite brings significant technical upgrades, including an expanded context length of 128K tokens and support for eight different languages. Meta also provides new components that transform Llama into a comprehensive system, empowering developers to build custom agents, generate synthetic data, and perform model distillation with unprecedented flexibility and control.
To support this massive release, Meta bolsters the ecosystem with over 25 launch partners like AWS, NVIDIA, and Google Cloud, while introducing new safety tools such as Llama Guard 3 and Prompt Guard. The company also releases a request for comment on the Llama Stack API, aiming to establish a standard interface that makes it significantly easier for third-party projects to integrate these advanced models.