Meta is facing a significant exodus of employees associated with the company’s AI teams. Many key researchers behind the company’s open-source Llama models are leaving, raising concerns within the company. This includes several of the authors credited on the landmark 2023 paper that introduced Llama — only three of the fourteen remain at Meta. The rest of them have left and went on to found their own companies, or join rivals.
Several former Meta researchers have been lured away by Mistral, a French startup. Mistral’s co-founders Guillaume Lample and Timothee Lacroix are building powerful open-source models that directly compete with Meta’s flagship AI efforts.
READ: Former Meta VP Roi Tiger raises funds for his new startup (March 25, 2025)
Meta also saw the departure of Joelle Pineau, who led the company’s Fundamental AI Research group (FAIR) for eight years. She announced her resignation last month.
Meta has laid off thousands of employees in a bid to focus its efforts on the AI race, spending significant amounts of capital—as much as $72 billion this year alone—on technical infrastructure and talent. However not all of its efforts towards furthering AI are paying off. Llama 4, the company’s latest AI release, has been poorly received, and it’s largest-ever AI model, dubbed “Behemoth,” has reportedly been delayed due to internal concerns about its performance.
Meta also still lacks a dedicated “reasoning” model for multi-step thinking or problem-solving tasks. This has become a matter of concerns since rival companies like OpenAI and Google focus on these features in their latest models.
The 2023 Llama paper was a major technical milestone for Meta, since it legitimized open-weight large language models with freely available code and parameters. Meta trained its models on publicly available data and optimized them for efficiency, allowing researchers to run state-of-the-art systems on a single GPU chip. However, the company eventually lost its lead in the open-source AI space to competitors like DeepSeek.