Yu Jiahui Leaves Meta to Start New Venture, Led Muse Multimodal Models

On August 14, 2026, Yu Jiahui announced on social media that he was leaving Meta to start a new venture. His personal website has been updated with his resume. In his announcement, he said that working with Zuckerberg and Meta's Chief AI Officer Alexandr Wang to build TBD Lab was 'very inspiring and fulfilling,' and mentioned that he was increasingly drawn to a question that is 'important for humanity's future but not yet fully explored,' without revealing the new company's name, research direction, or partners.

Yu Jiahui joined Meta at the end of June 2025, along with OpenAI researchers Zhao Shengjia, Bi Shuchao, and Ren Hongyu, to join the newly formed superintelligence team. Zuckerberg subsequently announced the creation of Meta Superintelligence Labs (MSL), led by Alexandr Wang, with TBD Lab as its core model research team. As a founding member and head of multimodal direction, Yu led his team to release Muse Spark in April 2026, MSL's first foundation model, marking Meta's shift from Llama to the Muse series for its frontier models. The team later released Voice Mode, Muse Image, and Muse Video, covering voice, image, and video generation. Days before his departure, Meta released Muse Spark 1.2, focusing on coding, long-horizon agent tasks, and multimodal capabilities.

Yu Jiahui's compensation has been a focus of external attention. During last year's Silicon Valley talent war, Altman publicly stated that Meta offered OpenAI employees $100 million signing bonuses and higher annual salaries; WIRED reported that Meta offered top AI talent packages worth up to $300 million over four years, with first-year total compensation potentially exceeding $100 million. However, Meta CTO Andrew Bosworth clarified that the $100 million was not for everyone and was not a signing bonus paid upfront, but rather an overall compensation package including stock, bonuses, tenure, and performance conditions. Therefore, whether Yu actually received a '$100 million annual salary' remains unanswered.

Yu Jiahui's resume is top-tier in the AI field: he studied at the Special Class for the Gifted Young of the University of Science and Technology of China, then earned his PhD under Thomas Huang at UIUC, where he contributed to DeepFill. He worked at Google Brain, co-leading the Gemini multimodal direction, and in late 2023 moved to OpenAI to lead the Perception team, contributing to GPT-4o, GPT-4.1, and the o-series models. At the time, Meta was in urgent need of rebuilding its frontier model capabilities after Llama 4 underperformed, and Yu's experience in vision research, large model development, and productization fit the need perfectly.

After Yu Jiahui's departure, the alphaXiv platform published a list of former Meta researchers, claiming that over 200 prominent researchers had left Meta before Yu, with its search page showing 929 researchers matching the criteria of 'previously employed at Meta but not currently.' The list also included researchers who moved to Google, became chief scientists at startups, or joined stealth projects. alphaXiv quipped: 'The future of Meta AI is also TBD (To Be Determined).'

This data has raised questions about Meta's talent strategy. Last year, when Meta was waving checks to poach talent, Altman had said, 'Missionaries will eventually defeat mercenaries.' Now that Yu, who was heavily recruited, has also left, the outside world is debating whether Meta's 'mercenary' strategy has succeeded. For Yu, the salary, compute, and product opportunities that a big company can offer may have approached the ceiling for an employee, making entrepreneurship a natural choice.