Meta FAIR

Type: Research Lab

Meta's Fundamental AI Research (FAIR) lab is the home of Yann LeCun's Advanced Machine Intelligence (AMI) program. FAIR occupies a distinctive position in the AI landscape: it produces frontier research and releases models as open weights, including the LLaMA series, the JEPA family (I-JEPA, V-JEPA, V-JEPA 2), and a steady stream of computer-vision and multimodal research.

Research Focus

• World models via JEPA — the flagship program; V-JEPA 2 is the headline 2025 result
• Open-weight foundation models — LLaMA (not covered in this KB section)
• Self-supervised learning for vision and video — DINOv2, I-JEPA, V-JEPA lineage
• Robotics via learned world models — V-JEPA 2-AC zero-shot Franka pick-and-place

Key Releases

• V-JEPA 2 (Jun 2025) — 1M+ hours of video pre-training; 77.3% on Something-Something v2; SOTA on Epic-Kitchens-100 anticipation; zero-shot robotic manipulation (paper)
• V-JEPA (2024) — the original video JEPA
• I-JEPA (2023) — image JEPA, the first successful JEPA at scale
• LLaMA series — open-weight LLMs (not covered here)

Positioning vs. Other Labs

• Vs. Google DeepMind: FAIR represents the JEPA / abstract-prediction camp; DeepMind's Genie series represents the generative / pixel-prediction camp. Direct architectural disagreement at the frontier.
• Vs. OpenAI / Anthropic: FAIR's open-weight stance contrasts sharply with both labs' closed-weight proprietary approach. LeCun has been vocal that closed-weight labs advocate for AI regulation as competitive moat.

Mentioned In

• World Models
• JEPA

Related Entities

• Yann LeCun — VP & Chief AI Scientist
• Google DeepMind — primary architectural rival on world models
• OpenAI
• Anthropic

Changelog

• 2026-04-22 — Initial page. Focus on world-models research; LLaMA lineage deferred.