Meta AI is the artificial intelligence research and product organization of Meta Platforms (formerly Facebook), headquartered in Menlo Park, California. It traces its research lineage to Facebook AI Research, a lab founded in 2013 under the founding direction of Yann LeCun, and became widely known during the 2020s generative AI wave for its Llama family of large language models, released with unusually permissive access to their weights relative to rivals such as OpenAI and Anthropic.
History
Facebook AI Research (FAIR) was established in 2013 as one of the first industrial labs dedicated to fundamental AI research rather than only applied product work, and under LeCun, whose earlier career included foundational Convolutional neural network research, it built a reputation for open publication and released tools including the PyTorch deep learning framework, which became one of the two dominant frameworks in the field. As generative AI shifted from a research curiosity to a central corporate priority following the 2022 launch of ChatGPT, Meta reorganized its AI efforts, elevating language model development to a company-wide priority under chief executive Mark Zuckerberg and folding FAIR into a broader Meta AI product and research umbrella.
Open-weights strategy
Meta's most consequential strategic choice was to release its Llama models with weights available for download and, from Llama 2 onward, for commercial use under Meta's own license, contrasting with OpenAI's, Anthropic's, and Google's practice of serving models only through hosted APIs. Meta framed this as both a commitment to broad access and a competitive strategy: by seeding an ecosystem of developers, researchers, and companies building on Llama, Meta aimed to make its models a default choice for Open-weights models deployment even without matching the very largest closed models on every benchmark. Llama's first version leaked ahead of its intended limited release in early 2023, accelerating public interest, and subsequent versions were released with expanding capability and, eventually, deliberate open publication, reinforcing Meta's position as the leading well-resourced publisher of openly available Foundation models.
2025 reorganization
In 2025, facing intensifying competition and reports of internal setbacks with a planned flagship model, Meta restructured its AI efforts again, forming a new "superintelligence" lab and recruiting senior researchers and executives from rival companies, including a high-profile hire of Alexandr Wang from Scale AI into a senior AI leadership role at Meta, alongside substantial compensation packages reported for other recruits, reflecting the intensity of talent competition among frontier labs during this period.
Reception
Meta AI's open-weights approach has been widely credited with expanding access to capable models for researchers, startups, and countries without the resources to train frontier systems from scratch, and with pressuring closed labs on price and features. Critics have raised concerns that openly released weights make safety-relevant restrictions easier to remove and could ease misuse, a tension Meta has addressed through licensing terms and acceptable-use policies of debated effectiveness. The 2025 reorganization drew further scrutiny over whether Meta's historically research-driven, LeCun-associated culture would persist under more product- and talent-acquisition-driven leadership.