TellWorldAI World Archive
Ask中文
News
Models
Organisations
People
Papers & concepts
Products & agents
Chips & robots
History
Topics
Ask中文

Home›Companies

OrganizationResearch lab

Meta AI

Meta AI is the AI research and model-building organisation of Meta Platforms. Founded in late 2013 as Facebook AI Research, it produced PyTorch and the Llama and Muse model families, and it is separate from the Meta AI assistant product.

Founded 2013Website ai.meta.com Era Deep learning

2024 · 9 20132026
Events per year
23Models & products33Events10People1Investors & partners

What it is

Meta AI is the artificial intelligence research and model-building organisation inside Meta Platforms, the company formerly known as Facebook. The name also covers Meta's AI assistant, which is a product built by the organisation; this entry covers the organisation and its research and models.[1]

Origins

The organisation was founded in late 2013 as Facebook AI Research (FAIR). Mark Zuckerberg went to the NeurIPS conference to recruit researchers, and partnered with Yann LeCun to build the founding team.[1] LeCun became FAIR's first director on 9 December 2013, in New York City. He was succeeded as director by Jérôme Pesenti in 2018, and stayed on as chief AI scientist.[2][3]

After Facebook rebranded as Meta Platforms in 2021, FAIR continued as Meta's Fundamental AI Research team. Meta has described open research as its founding purpose.[3][1]

Key work and releases

Research highlights listed by Meta include Faster R-CNN (2015), Mask R-CNN (2017), Segment Anything (2023), and No Language Left Behind, a translation and speech technology covering more than 1,000 languages.[1]

PyTorch, the deep-learning framework, began development in 2016 with people and funding from Meta and the Lua Torch community. It was publicly released in January 2017 and moved to the Linux Foundation as the PyTorch Foundation on 12 September 2022.[4][5]

In language models, Meta released OPT-175B in May 2022 under a research license. Galactica followed in November 2022, but its demo was withdrawn after three days. LLaMA came on 24 February 2023, and Llama 2 on 18 July 2023, free for research and commercial use. Code Llama followed in August 2023.[6][7][8]

Llama 3 arrived on 18 April 2024, pretrained on over 15 trillion tokens. Llama 3.1 405B followed on 23 July 2024. It was trained on over 16,000 H100 GPUs, and Meta described it as the largest openly available foundation model. Llama 3.2 in September 2024 added vision and on-device models.[9][10][11]

Llama 4 arrived on 5 April 2025. It introduced mixture-of-experts models that are natively multimodal: Scout, with 109 billion total parameters and a 10-million-token context, and Maverick, with 400 billion total parameters. A roughly 2-trillion-parameter model called Behemoth was previewed but never released.[12][13]

Meta reported that Llama models passed 300 million downloads by July 2024 and one billion by April 2025.[10][14]

Products

The Meta AI assistant launched in beta in the US on 27 September 2023. By June 2025 Meta said it had more than 1 billion monthly active users across its apps.[15][16]

Superintelligence Labs

In June 2025 Meta invested $14.3 billion in Scale AI for a 49% non-voting stake, and Scale's chief executive Alexandr Wang joined Meta.[17]

On 30 June 2025 Mark Zuckerberg announced Meta Superintelligence Labs (MSL), led by Wang as chief AI officer, with Nat Friedman heading AI products. MSL combined the foundations, product and FAIR teams with a new lab for next-generation models.[16]

In August 2025 MSL was split into four groups: TBD Lab for large models, FAIR, Products and Applied Research, and MSL Infra. Yann LeCun announced on 19 November 2025 that he was leaving to found a startup.[18][2]

MSL's first model, Muse Spark, launched on 8 April 2026 as a closed model powering Meta AI. Muse Image and Muse Video followed in July 2026, and the open-weight Muse Spark Open, with 816 billion total parameters, arrived on 15 October 2026.[19][20][21]

Where it fits

Meta AI is behind several important open AI tools: the PyTorch framework, the Llama models that developers have downloaded in large numbers, and the Muse models, which now come in both closed and open forms. The history runs from research papers and open releases to large model families tied to products. Whether each model is open or closed is decided model by model.

CapabilitiesLarge language models

TechniquesFoundation models

Ecosystem & regionOpen sourceDeveloper tools

HistoryResearch history

20265

  1. AcquisitionMeta acquires AI coding startup Conductor

    In early November 2026 Meta acquired Conductor, a startup whose platform runs multiple AI coding agents in parallel workspaces. Conductor's cloud service was set to shut down on 20 November 2026.

  2. ReleaseMeta releases the open-weight Muse Spark Open

    On 15 October 2026 Meta released Muse Spark Open, an open-weight, natively multimodal model with 816 billion total parameters, optimised to run on Apple's Mac Studio. Its weights were published on Hugging Face.

  3. UpdateMeta releases Muse Spark 1.1 and opens the Meta Model API

    On 9 July 2026 Meta released Muse Spark 1.1, an upgrade built for agentic tasks, with a 1-million-token context and access for developers through the Meta Model API in public preview.

  4. ReleaseMeta introduces Muse Image and Muse Video

    On 7 July 2026 Meta Superintelligence Labs introduced Muse Image and Muse Video, its first media generation models. Muse Image launched in Meta's apps, while Muse Video entered early preview.

  5. ReleaseMeta Superintelligence Labs releases Muse Spark

    On 8 April 2026 Meta Superintelligence Labs released Muse Spark, the first model from the lab and the first of the Muse series. It is closed: its weights were not released, and it was offered through a private API preview.

20258

  1. DepartureYann LeCun leaves Meta to found an AI startup

    On 19 November 2025 Yann LeCun confirmed that he would leave Meta after ten years to found an AI startup focused on advanced machine intelligence. Meta was to be a partner of the new company.

  2. ReleaseMeta releases SAM 3 and SAM 3D

    On 19 November 2025 Meta released SAM 3, which detects, segments and tracks objects from text or visual prompts, and SAM 3D, which reconstructs 3D objects and human bodies from a single image. Both were released as open models.

  3. PartnershipMeta licenses Midjourney's image and video technology

    On 22 August 2025, Meta said it would license Midjourney's AI image and video generation technology, with Meta research teams collaborating with the startup. Financial terms were not disclosed, and Midjourney said it remains independent.

  4. DeprecationMeta shuts down Papers with Code

    In July 2025 Meta shut down Papers with Code without notice, taking its 9,327 benchmark leaderboards and tens of thousands of paper-code links offline; the community rescued data snapshots, and Hugging Face's Trending Papers was named the replacement.

  5. MilestoneZuckerberg creates Meta Superintelligence Labs

    On 30 June 2025 Mark Zuckerberg announced Meta Superintelligence Labs (MSL), combining several AI teams under Alexandr Wang. Its first model, Muse Spark, was released on 8 April 2026.

  6. FundingMeta invests $14.3 billion in Scale AI and hires Alexandr Wang

    On 12 June 2025 Meta contributed $14.3 billion to Scale AI for a 49% stake without voting power. Scale AI founder Alexandr Wang left to join Meta's superintelligence efforts and was later named chief AI officer.

  7. ReleaseMeta launches a standalone Meta AI app

    On 29 April 2025 Meta launched a standalone Meta AI app for iOS and Android, built on Llama 4. It added a feed of shared prompts, voice conversations and personalised answers based on profile information.

  8. ReleaseMeta releases Llama 4 Scout and Maverick

    On 5 April 2025 Meta released Llama 4 Scout and Maverick, its first mixture-of-experts and natively multimodal Llama models. Meta also previewed the larger Behemoth model, which was still in training.

20249

  1. ReleaseMeta releases Llama 3.3 70B

    On 6 December 2024 Meta released Llama 3.3, a 70B text model that Meta said performs similarly to Llama 3.1 405B at much lower cost. It was the last Llama 3 release before Llama 4.

  2. MilestoneMeta announces its largest data center in Richland Parish, Louisiana (Hyperion)

    In December 2024 Meta announced a 4-million-square-foot, $10 billion-plus data center in Richland Parish, Louisiana, its 23rd US data center, built to train future open-source large language models and later expanded substantially.

  3. ReleaseMeta unveils Movie Gen media generation models

    On 4 October 2024 Meta premiered Movie Gen, a set of media foundation models that generate video, images and sound. Meta did not release the models, saying it would first work with filmmakers and creators.

  4. ReleaseMeta releases Llama 3.2 with vision and on-device models

    On 25 September 2024 Meta released Llama 3.2, which added image understanding to the Llama family with 11B and 90B vision models and added small 1B and 3B text models for phones and other edge devices.

  5. ReleaseMeta releases SAM 2 for video and image segmentation

    On 29 July 2024 Meta released SAM 2, one model that segments objects in both images and videos from simple prompts, running in real time. Its code and weights were released under open licences.

  6. ReleaseMeta releases Llama 3.1 405B, an open frontier-scale model

    On 23 July 2024 Meta released Llama 3.1 in 8B, 70B and 405B sizes. Meta called the 405B model the world's largest and most capable openly available foundation model, trained on over 15 trillion tokens with more than 16,000 H100 GPUs.

  7. FundingScale AI closes a $1 billion Series F at a $13.8 billion valuation

    In May 2024 Scale AI closed a $1 billion financing at a $13.8 billion valuation, led by existing investor Accel, with participation from existing backers such as NVIDIA and new ones including Amazon, Meta and Intel Capital.

  8. ReleaseMeta releases Llama 3

    On 18 April 2024 Meta released Llama 3 in 8B and 70B sizes, both pretrained and instruction-tuned. The models were pretrained on over 15 trillion tokens and made available through major cloud providers and Hugging Face under the Meta Llama 3 Community License.

  9. ReleaseNVIDIA announces the Blackwell platform

    On 18 March 2024 NVIDIA announced Blackwell, the GPU platform succeeding Hopper, claiming it can run real-time generative AI on trillion-parameter models at up to 25 times lower cost and energy.

20236

  1. ReleaseMeta introduces the Meta AI assistant and AI characters

    At Meta Connect on 27 September 2023 Meta introduced Meta AI, a conversational assistant in US beta on WhatsApp, Messenger and Instagram, along with 28 AI characters, 14 of them played by celebrities and influencers. It was rebuilt on Llama 3 in April 2024.

  2. ReleaseMeta releases Code Llama

    On 24 August 2023 Meta released Code Llama, a family of code models built by further training Llama 2 on code. It came in 7B, 13B and 34B sizes, under the same community licence as Llama 2, and a 70B version followed in January 2024.

  3. PolicySeven AI companies make voluntary commitments to the White House

    On 21 July 2023 President Biden convened seven leading AI companies at the White House and secured voluntary commitments covering eight measures, from watermarking AI-generated audio and video to independent safety testing before public release.

  4. ReleaseMeta releases Llama 2 for research and commercial use

    On 18 July 2023 Meta released Llama 2, a family of 7B, 13B and 70B parameter models free for research and commercial use under a custom license. Microsoft Azure, AWS and Hugging Face were among the platforms offering it.

  5. ReleaseMeta releases the Segment Anything Model and SA-1B dataset

    On 5 April 2023 Meta released the Segment Anything Model (SAM), a promptable image segmentation model, together with the SA-1B dataset of over 1.1 billion masks. Meta presented it as a step toward a foundation model for image segmentation.

  6. ReleaseMeta releases LLaMA to researchers

    On 24 February 2023 Meta announced LLaMA, a family of foundation language models with 7 to 65 billion parameters. Meta released it to approved researchers under a noncommercial license, and the paper described training on public data only.

20223

  1. ReleaseMeta launches Galactica and pulls its demo three days later

    On 15 November 2022 Meta AI and Papers with Code unveiled Galactica, a language model meant to help scientists, with a public demo. Meta took the demo offline about three days later after users showed it producing fabricated, confident answers.

  2. MilestoneMeta moves PyTorch to the new PyTorch Foundation

    On 12 September 2022 Meta announced that PyTorch would move to the Linux Foundation as the PyTorch Foundation, with a governing board that included AMD, AWS, Google Cloud, Microsoft Azure and NVIDIA. PyTorch stayed under the BSD-3-Clause licence.

  3. ReleaseMeta releases OPT-175B to researchers

    On 3 May 2022 Meta AI announced OPT-175B, a 175-billion-parameter language model, and released its pretrained weights and training code to researchers under a noncommercial licence. Meta said it was the first release of a model of that size with both weights and code.

20171

  1. ReleasePyTorch is publicly released

    PyTorch was publicly released in January 2017 as a beta, with first-class Python integration and an imperative style. Built by people from the Lua Torch community with Facebook's backing, it later became a widely used deep-learning framework.

20131

  1. MilestoneFacebook founds FAIR with Yann LeCun as director

    In late 2013 Facebook launched Facebook AI Research (FAIR), with Yann LeCun as its first director. The lab's open research model shaped Meta's later AI work, including PyTorch and the Llama models.

Models and products over time

23
  1. 2017
    • Jan 2017 PyTorch Product

      PyTorch is an open-source deep-learning framework released by Meta in 2017, now governed by the PyTorch Foundation under the Linux Foundation. It is used both for research and for training and running large models in production.

  2. 2022
    • May 3, 2022 OPT-175B (Open Pre-trained Transformer) Model

      OPT-175B is a 175-billion-parameter language model from Meta AI, announced in May 2022. Meta said it was the first model of its size released with both pretrained weights and the code to train and use them.

    • Nov 15, 2022 Galactica Model

      Galactica is a large language model for science, unveiled by Meta AI and Papers with Code in November 2022. Its public demo was taken down about three days later, after users found it produced fake papers.

  3. 2023
    • Feb 24, 2023 LLaMA Model

      A family of foundation language models from Meta AI, released in 7B to 65B parameter sizes and trained on public data; its 13B model was reported to beat GPT-3 on most benchmarks.

    • Apr 5, 2023 Segment Anything Model (SAM) Model

      The Segment Anything Model (SAM) is a promptable image segmentation model Meta AI released in April 2023. Given a point, a box or a mask, it outlines the object the prompt refers to.

    • May 2023 Meta MTIA v1 Chip

      MTIA v1 is Meta's first-generation in-house AI inference accelerator, built on TSMC's 7 nm process with 128 MB of on-chip memory and designed for deep learning recommendation models.

    • Jul 18, 2023 Llama 2 Model

      Meta's family of open-weight language models released in July 2023 in 7B, 13B and 70B sizes, each with a dialogue-tuned Chat variant, issued under a custom licence that allowed free research and commercial use.

    • Aug 24, 2023 Code Llama Model

      Code Llama is a family of coding models that Meta released in August 2023 by further training Llama 2 on code. It ranges from 7 billion to 70 billion parameters and uses the Llama 2 community license.

    • Sep 27, 2023 Meta AI (assistant) Product

      Meta AI is Meta's conversational assistant, introduced in US beta in September 2023 on WhatsApp, Messenger and Instagram. It was rebuilt on Llama 3 in 2024 and on Muse Spark in 2026.

  4. 2024
    • Apr 10, 2024 Meta MTIA v2 Chip

      Meta's second-generation in-house AI chip, published on 10 April 2024, built on TSMC's 5-nanometer process and aimed at the ranking and recommendation models Meta runs in its own data centers.

    • Apr 18, 2024 Llama 3 Model

      Llama 3 is an open model family from Meta, released on 18 April 2024 at 8 billion and 70 billion parameters. It was pretrained on over 15 trillion tokens of publicly available data.

    • Jul 23, 2024 Llama 3.1 (including 405B) Model

      Llama 3.1, released on 23 July 2024, added a 405-billion-parameter model to the Llama 3 family and extended its context to 128,000 tokens. Meta called the 405B the largest and most capable openly available foundation model.

    • Jul 29, 2024 SAM 2 (Segment Anything Model 2) Model

      SAM 2 is Meta's 2024 successor to the Segment Anything Model. It segments objects in both images and videos in real time, and its release included the SA-V video dataset.

    • Sep 25, 2024 Llama 3.2 Model

      Llama 3.2, announced on 25 September 2024, added vision models at 11 billion and 90 billion parameters and small text-only models at 1 billion and 3 billion for edge and mobile devices.

    • Oct 4, 2024 Movie Gen Model

      Movie Gen is a set of media foundation models Meta premiered on 4 October 2024. It covers video generation, personalised video, precise video editing and audio generation, and was not released publicly.

    • Dec 6, 2024 Llama 3.3 70B Model

      Llama 3.3 70B is a text-only, instruction-tuned model Meta released on 6 December 2024. Meta said it performs close to the 405B Llama 3.1 at much lower cost and can run on a typical developer workstation.

  5. 2025
    • Apr 5, 2025 Llama 4 Scout Model

      Llama 4 Scout is a mixture-of-experts model Meta released on 5 April 2025. It has 109 billion total parameters, 17 billion of them active for each token, and a context window of 10 million tokens.

    • Apr 5, 2025 Llama 4 Maverick Model

      Llama 4 Maverick is a mixture-of-experts model Meta released alongside Scout on 5 April 2025, with 400 billion total parameters and a 1-million-token context. Its ranking on LMArena was later questioned.

    • Apr 5, 2025 Llama 4 Behemoth Model

      Llama 4 Behemoth is a very large Meta model previewed on 5 April 2025 as the teacher for the Llama 4 family. It was never publicly released.

    • Nov 19, 2025 SAM 3 (Segment Anything Model 3) Model

      SAM 3 is Meta's segmentation model from November 2025. It finds, segments and tracks every instance of a concept named by a short text prompt or an example image, and Meta reported it doubles the accuracy of existing systems on this task.

  6. 2026
    • Apr 8, 2026 Muse Spark Model

      Muse Spark is the first model from Meta Superintelligence Labs, released on 8 April 2026 as a closed model that powers the Meta AI app and meta.ai. An open-weight sibling followed later.

    • Jul 7, 2026 Muse Image and Muse Video Model

      Muse Image and Muse Video are Meta Superintelligence Labs' first media generation models, announced on 7 July 2026. Muse Image launched immediately, and Muse Video entered early preview.

    • Oct 15, 2026 Muse Spark Open Model

      Muse Spark Open is Meta's open-weight large language model, announced on 15 October 2026. With 816 billion total parameters, it is built for coding and agentic work, and its weights can be downloaded, unlike the closed Muse Spark that started the family.

People10

Yann LeCun2013–2025Tomáš Mikolov2014–Kaiming He2016–2024Noam BrownMark ZuckerbergGuillaume Lample–2023Timothée Lacroix–2023Alexandr Wang2025–Daniel Gross2025–Nat Friedman2025–

Invested in1

Scale AI

Contributed to4

Notes and sources

  1. Celebrating 10 years of FAIR: A decade of advancing the state-of-the-art through open researchhttps://ai.meta.com/blog/fair-10-year-anniversary-open-science-meta/
  2. Yann LeCunhttps://en.wikipedia.org/wiki/Yann_LeCun
  3. Meta AIhttps://en.wikipedia.org/wiki/Meta_AI
  4. PyTorch is joining the Linux Foundationhttps://pytorch.org/blog/PyTorchfoundation/
  5. Meta hands PyTorch to the Linux Foundationhttps://www.theregister.com/2022/09/12/pytorch_meta_linux_foundation/
  6. Democratizing access to large-scale language models with OPT-175Bhttps://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/
  7. Why Meta's latest large language model survived only three days onlinehttps://www.technologyreview.com/2022/11/18/1063487/meta-large-language-model-ai-only-survived-three-days-gpt-3-science/
  8. Introducing Code Llama, a state-of-the-art large language model for codinghttps://ai.meta.com/blog/code-llama-large-language-model-coding/
  9. Introducing Meta Llama 3https://ai.meta.com/blog/meta-llama-3/
  10. Llama 3.1https://ai.meta.com/blog/meta-llama-3-1/
  11. Llama 3.2: Vision and edge deviceshttps://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/
  12. The Llama 4 herd: The beginning of a new era of natively multimodal AI innovationhttps://ai.meta.com/blog/llama-4-multimodal-intelligence/
  13. Meta's new superintelligence lab discusses major AI strategy changes, including abandoning Behemothhttps://gigazine.net/gsc_news/en/20250715-meta-built-ai-strategy-changes
  14. Everything we announced at our first-ever LlamaConhttps://ai.meta.com/blog/llamacon-llama-news/
  15. Introducing New AI Experiences Across Our Family of Apps and Deviceshttps://about.fb.com/news/2023/09/introducing-ai-powered-assistants-characters-and-creative-tools/
  16. Mark Zuckerberg announces creation of Meta Superintelligence Labs. Read the memohttps://www.cnbc.com/2025/06/30/mark-zuckerberg-creating-meta-superintelligence-labs-read-the-memo.html
  17. Scale AI founder Wang announces exit for Meta as part of $14 billion dealhttps://www.cnbc.com/2025/06/12/scale-ai-founder-wang-announces-exit-for-meta-part-of-14-billion-deal.html
  18. Meta Superintelligence Labshttps://en.wikipedia.org/wiki/Meta_Superintelligence_Labs
  19. Introducing Muse Spark: Scaling Towards Personal Superintelligencehttps://ai.meta.com/blog/introducing-muse-spark-msl/
  20. Introducing Muse Image and Muse Videohttps://ai.meta.com/blog/introducing-muse-image-muse-video-msl/
  21. Meta launches new Muse Spark Open model, optimized to run on Mac Studiohttps://9to5mac.com/2026/10/15/meta-launches-new-muse-spark-open-model-optimized-to-run-on-mac-studio/

Cite this page: Meta AI. TellWorld, https://www.tellworld.net/companies/meta-ai. Updated 2026-11-08