Meta’s Muse AI Agent, Kept Under Wraps for Half a Year, Teams Up with NVIDIA to Scale Up
According to the latest report from tech leak site TestingCatalog, NVIDIA will expand its collaboration with Meta on the Muse project, focusing primarily on AI agent development. The two companies have already cooperated deeply on Meta’s Llama series of large language models. As of now, neither company has officially confirmed the news.
Tech media outlet TestingCatalog, which specializes in uncovering unreleased features from major tech firms, recently broke the news that NVIDIA and Meta will expand their cooperation scope on the Muse project, with the new round of collaboration focusing mainly on AI agents. The article is written by Erin from the site’s AI Agent section. As of press time, neither company has responded officially to the news.
People who follow early tech product developments are no strangers to TestingCatalog. Their information comes largely from beta app disassembly, code residue mining, and internal source verification. Over the past two years, they have scooped many news ahead of time, such as Meta AI’s feature updates and Instagram’s generative tool launch. Most of the content they broke has later been verified, so their credibility ranks among the top tier in the tech leak community.
Muse is Meta’s internal next-generation AI assistant project, with relevant clues first leaked in mid-2024. Unlike Meta’s current Meta AI, which is a question-and-answer type assistant integrated into Facebook, Instagram, and WhatsApp, Muse is positioned as a task-executing AI agent, aiming to complete specific operations across Meta’s ecosystem apps—such as organizing album categories, replying to messages in batches, filtering product information, and booking services—without requiring users to issue step-by-step instructions. The project is still in internal testing and has no public launch timetable yet.
NVIDIA and Meta have long been deep partners in the generative AI field. Starting with Llama 2, Meta’s large language model training has been fully based on NVIDIA’s GPU clusters, and the two sides have already established a complete process in computing power adaptation, inference optimization, and open-source ecosystem co-construction. The Llama 3 launched in 2024 also has support from NVIDIA’s technical team.
So far, the leaked content has not mentioned specific details of this cooperation. Whether NVIDIA will provide dedicated computing power support for Muse, or integrate its own AI agent development stack and simulation testing tools with Muse’s underlying framework, there is no clear information for now.
Many industry views previously held that Meta, with the world’s most mature open-source large model ecosystem, would likely adopt a full-stack self-research route for AI agents. Now, its initiative to deepen cooperation with NVIDIA also proves indirectly that the engineering complexity of implementing general-purpose AI agents is far higher than that of training a large model with tens of billions of parameters. Even top-tier tech giants with sufficient manpower and computing power reserves do not want to step on all the pits alone.
发布时间: 2026-10-27 19:35