Meta AI is Meta Platforms' AI division, based in Menlo Park, California since its founding as FAIR. It builds the open-weight Llama family and the closed Muse line, the newest being Muse Wonder, announced October 12, 2026, with a 1-million-token context window. Developers reach Muse through the Meta Model API and Muse Code.
Meta AI, which began as Facebook AI Research (FAIR), is Meta Platforms' AI division and maker of the open-weight Llama family (1 billion-plus downloads) and the closed Muse models. The newest model, Muse Wonder, has a 1-million-token context window and costs $4.00 input and $20.00 output per million tokens on the Meta Model API.
Founded: 2013 · HQ: Menlo Park, California, USA · Team: 70,000+ companywide (Meta Platforms, post-May 2026 layoffs); Meta AI/MSL headcount not separately disclosed · CEO: Mark Zuckerberg (Chairman & CEO) · Funding: Public (NASDAQ: META); $14.3B for 49% non-voting stake in Scale AI (Jun 2025); 2026 AI capex guidance raised to $125B-$145B (up ~87% YoY from $72.2B in 2025) · Valuation: $1.4T+ (Meta Platforms market cap, mid-2026)
About Meta AI
Meta AI traces to April 2013, when it launched as Facebook Artificial Intelligence Research (FAIR) under Meta (then Facebook) CEO Mark Zuckerberg, with Yann LeCun recruited from NYU as founding director. Facebook, Inc. renamed itself Meta Platforms, Inc. in October 2021. FAIR was never a separate legal entity; it operates as an internal division of the Nasdaq-listed parent, headquartered at Meta's Menlo Park, California campus.
Meta AI ships two model tracks. Llama is an open-weight family that began with Llama 1 in February 2023 and has passed 1 billion cumulative downloads across Llama 2, Llama 3 and Llama 4. Llama 4, released April 2025, used a mixture-of-experts architecture; a larger Behemoth variant was delayed indefinitely after researchers alleged Meta had gamed public benchmark leaderboards. The second track is the closed Muse family, built by the Meta Superintelligence Labs (MSL). Muse Spark launched April 8, 2026 and Muse Spark 1.1 followed on July 9, 2026 with a paid Meta Model API. Meta's developer site now lists Muse Spark 1.3 (released September 2, 2026 on Muse Code and the Meta Model API, with max reasoning), Muse Spark 1.2 and 1.1, plus Muse Glimmer for local always-on agents on Hugging Face, Muse Image, Muse Voice Transcribe and the Segment Anything model SAM 3.1.
The newest step is the Muse Wonder series, announced October 12, 2026. Muse Wonder is available through the Meta Model API and Muse Code with a 1-million-token context window, at $4.00 per million input tokens, $0.20 per million cached input tokens and $20.00 per million output tokens on the developer page. Muse Wonder Full is the more capable variant and is included only with a Muse Code subscription, on the Power Usage and Maximum Usage plans. Meta describes the series as built for personal and professional agents, pairing "Active Memory" (carrying context across conversations) with "Active Focus" (sustained work on hard goals), and says it was trained at the Prometheus data center. Meta says Muse Wonder will later power experiences in the Muse app, starting with subscribers. The Meta Model API itself is labelled public preview. Muse Code, Meta's multi-agent coding tool for the terminal, is tracked on Hokai as the agent Muse Code.
Meta AI does not raise venture capital; it runs on Meta Platforms' operating cash flow. For 2026, Meta raised its capex guidance to $125 billion to $145 billion, roughly double the $72.2 billion spent in 2025, driven mostly by data centers and custom silicon. In June 2025, Meta committed $14.3 billion for a 49 percent stake in Scale AI, bringing in co-founder Alexandr Wang; a planned $2 billion Manus acquisition was blocked by Chinese regulators in April 2026.
Meta AI generates no standalone revenue; results sit inside Meta Platforms, still over 98 percent advertising-driven. Q1 2026 revenue reached $56.31 billion, up 33 percent year over year. On July 1, 2026, Meta announced Meta Compute, a cloud business reselling excess AI capacity, with deals reported by CNBC at $1.25 billion per month with Anthropic and $920 million per month with Google.
Mark Zuckerberg remains Meta Platforms' CEO. Alexandr Wang, Scale AI's co-founder, joined as Chief AI Officer in mid-2025 and co-leads MSL with Nat Friedman, former GitHub CEO, while Shengjia Zhao, formerly of OpenAI, is chief scientist. MSL splits into four groups: TBD Lab for large language models, FAIR for research, Products and Applied Research, and MSL Infra. Founding FAIR director Yann LeCun left Meta in November 2025 and started the rival lab AMI Labs. Meta cut roughly 8,000 jobs in May 2026, about 10 percent of its workforce, while redirecting 7,000 employees into new AI units; headcount fell from 78,865 at the end of 2025 to an estimated 70,000.
MSL states its mission as building personal superintelligence safely and making it available to everyone. Since April 8, 2026, frontier releases are governed by the Advanced AI Scaling Framework version 2, gating deployment against chemical/biological, cybersecurity, and loss-of-control risks, and introducing published Safety and Preparedness Reports. FAIR continues publishing open research in computer vision, NLP, speech and reasoning, and remains the origin of the PyTorch framework.
Where Meta is weaker: it has no standalone enterprise trust center or published SOC 2 report for its AI products, so regulated buyers often find Anthropic further along on procurement paperwork, with OpenAI close behind. The search and Android distribution Meta lacks belongs to Google DeepMind. Meta's edge is free distribution: Meta AI sits inside Facebook, Instagram, WhatsApp and Messenger, plus the Ray-Ban AI glasses, and Llama weights remain available through hosts such as Together AI and Fireworks AI. Open-weight rivals fight for the developer mindshare Llama built: DeepSeek competes hardest on price, while European buyers weigh Mistral AI as a local alternative. Meta is funding two gigawatt-scale data center campuses, Hyperion in Louisiana and Prometheus in Ohio. Compare options in Smart Match before committing a workload to any one lab.
Mission
Build personal superintelligence safely and make it available to everyone.
Products
- Muse Wonder / Muse Wonder Full (Model / Developer API): https://dev.meta.ai/models/muse-wonder/
- Muse Spark 1.3 (Model / Developer API): https://dev.meta.ai/models/muse-spark/
- Muse Code (Developer tool): https://dev.meta.ai/
- Llama 4 (Open-weight model family): https://ai.meta.com/blog/llama4/
- Meta AI Assistant (Consumer product): https://ai.meta.com/
- Ray-Ban AI Glasses (Hardware): https://www.ray-ban.com/usa/meta-smart-glasses
- Meta Compute (Cloud / GPU infrastructure service): https://www.cnbc.com/2026/07/01/meta-stock-cloud-ai-compute.html
Meta AI Models on Hokai
Compliance
GDPR, CCPA
Links
Frequently Asked Questions
What is Meta AI and what do they build?
Meta AI is the artificial intelligence division of Meta Platforms, the parent of Facebook, Instagram, WhatsApp and Messenger. It builds Llama, an open-weight family that developers can download and self-host, and the closed Muse models from Meta Superintelligence Labs. The newest Muse releases are Muse Spark 1.3 and the Muse Wonder series, which Meta positions for personal and professional agents. Consumers meet the work through the Meta AI assistant inside the apps and through Ray-Ban AI glasses.
Who founded Meta AI and who is the CEO?
Meta AI began as FAIR under Mark Zuckerberg, who remains Chairman and CEO of Meta Platforms. Yann LeCun was founding director and left in November 2025 to start AMI Labs. Alexandr Wang, formerly Scale AI CEO, is Chief AI Officer and co-leads Meta Superintelligence Labs with Nat Friedman. Shengjia Zhao serves as chief scientist.
How much funding has Meta AI raised?
Meta AI has raised no outside funding because it is financed from Meta Platforms cash flow. Its largest external bet was 14.3 billion dollars for a 49 percent non-voting stake in Scale AI in June 2025. Meta guided 2026 capital spending to between 125 and 145 billion dollars, mostly data centers and custom chips. Parent-company results, not a venture valuation, are the only financial signal.
What products does Meta AI make?
Meta AI makes the open-weight Llama models, the closed Muse models, the Meta Model API and Muse Code, a multi-agent coding tool for the terminal. Meta lists Muse Spark 1.3, 1.2 and 1.1 alongside Muse Wonder and its Full variant, plus Muse Glimmer, Muse Image, Muse Voice Transcribe and SAM 3.1 on its developer site. Wonder Full comes bundled with higher Muse Code plans, while plain Muse Wonder is billed per token through the API. Spark 1.1 launched on the API in July 2026 at 1.25 dollars input and 4.25 dollars output per million tokens.
Is Meta AI compliant with SOC 2, GDPR, HIPAA?
Meta Platforms publishes a privacy policy and says it follows GDPR and CCPA for consumer AI surfaces, with an opt-out for EU and UK users whose public posts could train models. Meta has not published a SOC 2 Type II report, ISO 27001 certificate or HIPAA agreement for its AI API. Meta says Model API inputs are not used for training unless a customer opts in. Regulated buyers should treat the Model API, still in public preview, as less mature than dedicated enterprise vendors.
Top Alternatives
- OpenAI: Pick OpenAI if you need the strongest frontier reasoning scores; pick Meta AI if you want to self-host open weights for free.
- Google DeepMind: Pick Google DeepMind if you need native Search/Android/Workspace integration; pick Meta AI if you want a permissive open-weight license.
- Anthropic: Pick Anthropic if enterprise compliance depth matters most; pick Meta AI if you want an open-weight option alongside a closed Muse API.
- DeepSeek: Pick DeepSeek if you want the cheapest open-weight alternative; pick Meta AI (Llama) for the larger fine-tuning and hosting ecosystem.
Hokai guides that mention Meta AI
- Best AI Voice Recorders in 2026: What They Cost and What the Law Allows: Plaud, HiDock and Bee compared on two-year cost, where your audio goes and one-party versus all-party recording law, with every price read on 6 November 2026.
- Best Open-Weight Models in 2026: Ranked by Job, Hardware and Licence: The best open-weight models in 2026, ranked by job and hardware, with licence terms and Artificial Analysis scores read from each model card on 5 November.
- Personal Agent Protocol Explained: How AI Agents Will Sign In to Shops and Banks: Sierra and Meta's Personal Agent Protocol lets AI agents sign in to shops and banks. What the 9 October draft says, who gains, who loses, and what is missing.
- AI Agent Approval: What Personal AI Agents Can Do Without Asking You: Claude Duet, OpenAI dots, Meta Muse, Grok Bot, Poke, Perplexity Computer and OpenClaw: which actions wait for your approval and where your passwords live.
- Revnu vs Meta Muse for Small Business: A $99 Growth Hire or Meta's Free Agent?: Revnu costs $99 a month and runs Google, LinkedIn and cold outreach. Muse for Small Business is free and reads your Meta ads. Pricing, channels, which to pick.
- Best AI Marketing Agents in 2026: The Ones That Do the Work, Not Just Draft It: Revnu, Meta Muse, HubSpot Breeze and more AI marketing agents compared: what each can publish or spend without your approval, and what each costs in 2026.