← All Companies

Meta AI

πŸ“ United States πŸ“… Founded 2013 🏒 enterprise

Meta AI is the artificial intelligence research and products division of Meta Platforms (formerly Facebook). Its Chief AI Scientist is Yann LeCun, one of the three researchers awarded the 2018 Turing Award for deep learning. Meta has made a strategic bet on open-weight models: Llama 1, 2, 3, and 4 have all been released with publicly available weights, making Meta the dominant force in open-source large language models. This strategy contrasts sharply with OpenAI and Anthropic's closed approach and has reshaped the competitive landscape. AQCA notes that "open-weight" is not the same as "open-source" β€” access to weights does not provide access to training data or methodology. Meta's approach raises significant governance questions that the EU AI Act's GPAI provisions are still working through.

AI Models

Llama 3.3 70B

Open Source LLM

Released in December 2024, Llama 3.3 70B achieves performance comparable to the earlier Llama 3.1 405B model at less than one-fifth the parameter count. It is Meta's most capable open-weight model available at a size that can be run on a single high-end GPU. Released under Meta's Llama licence (not a standard open-source licence β€” commercial use above 700 million monthly users requires a separate agreement). Llama 3.3 70B is the dominant open-weight model for enterprise deployment as of early 2025.

Licence: Limited Size: 70B Released: Dec 2024
enes