Featured on
Foundation models · Menlo Park, United States
The clearest example of a Big Tech incumbent restructuring around the foundation-model race in real time - built by pulling in a rival's founder, and quietly abandoning the open-weight strategy that made Llama famous.
Meta Superintelligence Labs (MSL) was announced by Mark Zuckerberg on June 30, 2025, bringing together Meta's foundation-model development (the Llama model family), its long-running FAIR fundamental-research group, product integration work, and AI infrastructure engineering under one umbrella - a response to Meta's AI division falling behind OpenAI, Anthropic and Google on frontier model performance.
In August 2025, MSL restructured into four teams: TBD Lab, developing Llama models under Alexandr Wang (former CEO of Scale AI, who joined Meta as Chief AI Officer after Meta's investment in Scale); FAIR, Meta's fundamental research group, under Rob Fergus; Products and Applied Research, integrating models into Meta's consumer apps, under Nat Friedman; and MSL Infra, under Aparna Ramani. Its first major model under the new structure, Muse Spark, released in April 2026 as a closed-source system - a departure from Meta's earlier open-weight Llama strategy.
Llama
Meta's open-weight large language model family, widely used as a base for third-party fine-tuning.
Muse Spark
Natively multimodal reasoning model, released April 2026 as Meta's first closed-source flagship model.
Meta AI
Consumer AI assistant integrated across Meta's apps (Facebook, Instagram, WhatsApp).
meta.ai (opens in a new tab)Data-labeling and model-evaluation infrastructure company that trains and tests the models frontier labs build - now roughly half-owned by Meta.
The maker of ChatGPT and the GPT model family, and the company most responsible for the current wave of consumer generative AI.
AI safety-focused lab behind the Claude model family, founded by former OpenAI researchers.
