“The Llama 3 Herd of Models” was submitted to arXiv on July 31, 2024 by a large team at Meta led by Aaron Grattafiori and Abhimanyu Dubey, with hundreds of contributors. It documents the Llama 3 family of language models, which natively support multiple languages, coding, reasoning, and tool use. The flagship is a dense 405-billion-parameter model with a 128,000-token context window.
The paper reports that the 405B model delivers quality comparable to leading models such as GPT-4 across a wide range of tasks, making it one of the strongest openly released models at the time. Unusually for a frontier-scale release, Meta published a detailed technical account of the data, scaling decisions, and infrastructure behind the models, and released the weights openly. The paper also describes experiments integrating image, video, and speech capabilities through a compositional approach, with those multimodal extensions still under development.
The Llama 3 release was a landmark for open-weight AI: it narrowed the gap between freely downloadable models and the best proprietary systems, and its companion paper served as a rare, detailed playbook for how a model of that scale is actually built. For organizations, an openly available frontier-class model means the option to run, inspect, and customize top-tier AI without depending on a single vendor’s API.