Jarvis is the Artificial intelligence assistant developed by Meta's AI research division, Meta AI, and integrated into the company's social networking products, smartglasses, and virtual reality headsets. The assistant is built on Meta's Large language model family, Llama, and is available as a standalone app with optional paid tiers for higher usage limits. It represents Meta's push to embed generative AI capabilities across its consumer ecosystem, competing with assistants from OpenAI, Anthropic, and Google DeepMind.
The assistant traces its origins to Meta AI (formerly Facebook AI Research, or FAIR), founded in 2013 under the direction of Yann LeCun. FAIR's early work in Machine learning and Deep learning laid the groundwork for the models that now power Jarvis. Following the 2021 rebranding of Facebook, Inc. as Meta Platforms, FAIR continued as Meta's Fundamental AI Research team, and in 2025 the company consolidated its AI efforts under Meta Superintelligence Labs, which brought together FAIR, LLM development teams, and product groups.
History and Development
Meta AI was established in 2013 as Facebook AI Research (FAIR), with Yann LeCun as its founding director. LeCun led the organization until 2018, when Jérôme Pesenti took over. FAIR's research portfolio included self-supervised learning, generative adversarial networks, document classification, translation, and computer vision. In 2016, FAIR released fastText, an open-source library for efficient text classification and word representation learning. In 2017, the team released PyTorch, an open-source Neural network framework that became widely adopted across research and industry.
After the 2021 corporate rebranding, FAIR was renamed Meta's Fundamental AI Research team. In 2023, Meta introduced the Meta AI assistant, initially branded as such, based on the Llama model family. The assistant was integrated into Facebook, Instagram, WhatsApp, and Messenger. In 2025, Meta reorganized its AI efforts under Meta Superintelligence Labs, which unified FAIR, LLM development, and product teams. The assistant later became known as Jarvis, though Meta has not officially confirmed the name change as of early 2026.
Features and Integration
Jarvis is integrated into Meta's social networking products as a chatbot, accessible via direct messages and in-feed prompts. It is also available as a standalone mobile app, with optional paid subscriptions providing higher usage limits. The assistant is pre-installed on the second generation of Ray-Ban Meta smartglasses, where it can incorporate inputs from the glasses' cameras after a software update. It is also available on Quest 2 and newer virtual reality headsets, enabling voice-controlled interactions in immersive environments.
The assistant supports multimodal inputs, including text, voice, and images, and can perform tasks such as answering questions, generating text, summarizing content, and assisting with creative writing. It leverages Meta's Transformer (architecture)-based Llama models, which are trained on large-scale datasets and fine-tuned for conversational use. Jarvis also integrates with Meta's AI infrastructure, including the AI Research SuperCluster (RSC) and the Meta Training and Inference Accelerator (MTIA) custom chips.
Research and Models
Meta AI's research division has contributed significantly to natural language processing, including machine translation, natural language generation, and question answering. In 2023, the team released Seamless, a set of models for real-time voice translation. In 2024, researchers from Meta AI published work on machine translation for low-resource languages, with an editorial in Nature emphasizing the need for linguist involvement.
The Llama model family, which powers Jarvis, was first released in February 2023. As of January 2026, the most recent release is Llama 4, which introduced improvements in reasoning and efficiency. In April 2026, Meta Superintelligence Labs introduced Muse Spark, the first model in the Muse family, a multimodal reasoning model designed for tool use, visual reasoning, and multi-agent workflows. Meta subsequently expanded the family with Muse Image and Muse Video, which may inform future Jarvis capabilities.
Another notable model, Galactica, was an LLM designed for generating scientific text. It was available for three days from 15 November 2022 before being withdrawn due to generating racist and inaccurate content, highlighting the challenges of deploying AI systems without adequate safeguards.
Infrastructure and Hardware
Meta has developed large-scale computing infrastructure to train and deploy models like those powering Jarvis. In 2022, the company introduced the AI Research SuperCluster (RSC), a GPU-based supercomputer designed for training large models in NLP and computer vision. Meta also develops the MTIA family of custom AI accelerators for internal workloads, while continuing to use large clusters of Nvidia GPUs. This infrastructure supports the training of Llama models and the deployment of Jarvis across Meta's platforms, which serve billions of users worldwide.
Controversies and Ethical Concerns
Jarvis and Meta AI have faced several controversies. Since May 2024, the chatbot has summarized news from various outlets without linking directly to original articles, including in Canada, where news links are banned on Meta's platforms. This use of news content without compensation and attribution has raised ethical and legal concerns, especially as Meta continues to reduce news visibility on its platforms.
In 2022, the French media outlet Mediapart reported that Facebook's parent company illegally used works accumulated by the pirate site LibGen to train its AI models. This raised questions about copyright and the ethical sourcing of training data for models like Llama. Meta has not publicly confirmed or denied these allegations, but they remain a point of contention in the AI community.