Why Meta’s High Salaries Can’t Stop a Talent Exodus: Over 200 Top Researchers Depart

Meta Superintelligence Labs (MSL) core researcher Jiahui Yu announced on X that he is leaving Meta to found a new company, expressing pride in the multimodal achievements of Muse Spark, Voice Mode, Muse Image, and Muse Video.

Yu’s career has centered on multimodal AI: early work in computer vision and image generation, contributions to Google’s Parti and Gemini projects, and later leading OpenAI’s Perception team on GPT‑4o, o3, and o4‑mini. He joined Meta in June 2025, co‑building the TBD Lab with Mark Zuckerberg and Alexandr Wang.

In August, Meta released Muse Spark 1.2, which strengthens agent capabilities and multimodal interaction and powers new products such as Muse Code. At the same time, Muse Image launched, Muse Video is in progress, and Voice Mode was highlighted in Yu’s departure note.

Despite the recent model launch, Yu left after only 14 months at Meta, underscoring a broader talent churn.

According to the research‑paper platform alphaXiv, Meta has already lost more than 200 well‑known researchers. Analyst Gergely Orosz visualized an “iceberg” of Meta engineers: roughly 10% have been laid off, 20‑30% reassigned to tasks like data labeling and are mentally planning to leave, while the remaining 60‑70% stay but are actively submitting resumes.

Orosz cited a concrete example: an engineer reassigned to data labeling became upset, received a modest offer from Google, then a large counter‑offer from Meta, which Google matched and slightly exceeded. The engineer ultimately left Meta with a higher salary.

Orosz notes that among the seven engineers he knows who accepted counter‑offers, all eventually departed, with similar cases in Menlo Park, New York, and London. While Meta can close a model performance gap in a few months, rebuilding organizational trust will take considerably longer.