Meta AI
Open-Source AI
- Market Cap
- $2T+(Jun 2026)
- Revenue
- 2026 guided: $201B;3.58B daily active users across Meta apps
- Profit Status
- Highly profitable(FY2024 net income: $62B)
- Revenue Growth
- ~20% YoY
- Employees
- ~3,000 FAIR(Meta total: 74,000+)
About Meta AI
Meta AI is Meta's artificial intelligence research division. Champions open-source AI with the Llama model family and conducts fundamental research in NLP, computer vision, and robotics.
Company History
FAIR founded 2013. The open-weight Llama series set the global standard, ending with Llama 4 Scout and Maverick (Apr 2025), MoE with a 10M context; Llama 4 Behemoth was previewed but never publicly released. Meta then moved its frontier work to the proprietary Muse line: Muse Spark (8 Apr 2026) was its first closed-weight frontier model, built by Meta Superintelligence Labs rather than the Llama team, reaching 1.3 by Sep 2026. Yann LeCun left in Nov 2025 after twelve years. Meta AI passed 1B monthly active users (Oct 2025), and 2026 AI infrastructure capex guidance was raised to $125-145B.
AI Models & Benchmarks
7 modelsMuse Spark
Meta's first closed-weight, API-only frontier model. Shipped by Meta Superintelligence Labs (MSL). Marks major strategic shift toward proprietary AI.
Llama 4 Maverick
Best open-source multimodal model at launch. Native image understanding, 128-expert MoE.
Benchmarks
- MMLU
- 85.5%
- MATH
- 73.5%
- Multimodal
- #1 open at launch
Llama 4 Scout
World's longest context window (10M tokens) in open-source. Efficient 16-expert MoE.
Benchmarks
- MMLU
- 79.5%
- Context length
- 10M tokens - world record
Llama 3.3 70B
Best open-source text model value. Performance of Llama 3.1 405B at 70B size.
Benchmarks
- MMLU
- 86.0%
- HumanEval
- 88.0%
Llama 3.1 405B
First open-source frontier model - competitive with GPT-4o and Claude 3.5 Sonnet.
Benchmarks
- MMLU
- 88.6%
- HumanEval
- 89.0%
- MATH
- 73.8%
SAM 2
Segment Anything Model 2 - real-time segmentation for images and video.
SeamlessM4T v2
Nearly 100 languages. Speech-to-speech, speech-to-text, text-to-speech translation.
Company Timeline
9 milestones- Dec 2013FAIR founded, Yann LeCun appointed Chief AI Scientist
- Feb 2023LLaMA 1 released as open-weights model
- Jul 2023Llama 2 - commercially usable, sets open-source standard
- Jul 2024Llama 3.1 405B - first open-source frontier model
- Apr 2025Llama 4 Scout & Maverick at LlamaCon - MoE, native multimodal, 10M context
- Apr 2026Muse Spark (8 Apr) - Meta's first closed-weight frontier model, from Meta Superintelligence Labs
- Nov 2025Yann LeCun departs after twelve years; later founds AMI Labs
- Sep 2026Muse Spark 1.3 (2 Sep) - natively multimodal, 1M-token context
- 2026Meta AI passes 1B monthly active users (Oct 2025); 2026 AI capex guidance raised to $125-145B
Products & Services
6- Llama 3
- Llama 2
- SAM
- ImageBind
- Seamless
- Segment Anything
People & ecosystem
Key People
- Alexandr Wang (Chief AI Officer, Meta Superintelligence Labs)
- Shengjia Zhao (Chief AI Scientist, MSL)
Key Investors
- Public company (NASDAQ: META)