Meta Loses Key AI Researchers: Most Llama Creators Leave for Mistral AI

Meta Loses Key AI Researchers: Most Llama Creators Leave for Mistral AI

Ondřej Barták
Ondřej Barták
Entrepreneur and Programmer
28. 5. 2025
4 minutes reading
Meta Loses Key AI Researchers: Most Llama Creators Leave for Mistral AI

Meta Loses Key AI Researchers: Most Llama Creators Have Left for Mistral AI

Meta is facing a significant loss of top AI researchers who were instrumental in developing its foundational Llama language models. Of the 14 original authors of the groundbreaking Llama paper, only three remain at Meta, while most have left for competing companies or founded their own startups. The most prominent destination for these departures has been the Paris-based company Mistral AI, which was co-founded by former key architects of the Llama projects.

Massive Talent Exodus from Meta

The departure of researchers from Meta reveals a dramatic shift in the field of artificial intelligence. Of the original team of fourteen researchers who laid the foundations for the Llama models, only three now remain at Meta. This decline represents a major weakening of the company's innovation potential at a time when competition in generative AI is intensifying significantly. The most remarkable aspect of this exodus is the concentration of departing talent at Mistral AI. The French company was co-founded by three former key Meta employees: Guillaume Lample, Timothée Lacroix, and Arthur Mensch. After founding Mistral, they subsequently attracted other former colleagues from the Llama team, including Baptiste Rosier (Rozière), Marie-Anne Lachaux (Lachaud), and Thibaut Lavril.

Reasons for the Massive Departures

An analysis of the causes behind this talent loss reveals several factors that contributed to the researchers' decisions to leave Meta. One of the main reasons is competitive opportunities and the chance to join a startup environment that offers greater autonomy and influence over the direction of research. The environment at startups such as Mistral AI allows researchers to directly shape strategic decisions and work on projects without the bureaucratic obstacles of large corporations. Another significant factor is reports of innovation stagnation and delays within Meta. According to available information, development of the next-generation model known as "Llama 4 Behemoth" is facing problems and internal difficulties. These complications may have contributed to frustration among top talent seeking faster progress or more ambitious projects elsewhere. Broad corporate restructuring and uncertainty are also playing a role in researchers' decisions. Meta is undergoing extensive changes, including layoffs in divisions such as Reality Labs, which is facing significant operating losses. Such instability can undermine confidence in the company's long-term support for research.

Decline of Meta's Research Culture

Reports of changes in Meta's research culture are particularly concerning. Some insiders describe the once-renowned FAIR (Fundamental AI Research) lab as a dying institution. They primarily criticize rushed product releases, a lack of transparency, and possibly overstated claims about the performance of new models. These factors can erode researchers' morale and their confidence in management's decisions. Problems with the research culture are also reflected in the quality of outputs and the approach to scientific work. Pressure for quick results and commercial success may conflict with the rigorous scientific approach that top researchers require for their work.

Where the Key Researchers Went

The dispersion of talent from the original Llama team illustrates the wide range of opportunities available to top AI researchers. Guillaume Lample, Timothée Lacroix, and Arthur Mensch founded Mistral AI, where they serve as co-founders and key research scientists. Baptiste Rozière, Marie-Anne Lachaux, and Thibaut Lavril followed them to Mistral as scientists and research engineers. Other researchers from the original team found positions at other leading technology companies. Naman Goyal moved to a senior position at Thinking Machines Lab, while Aurélien Rodriguez became Director of Foundation Model Training at Cohere. Armand Joulin secured the prestigious position of Distinguished Scientist at Google DeepMind, Gautier Izacard moved to a technical role at Microsoft, and Edouard Grave became a research scientist at Kyutai.

Impact on Meta's Competitive Position

The departure of most of the original Llama team members signals a weakening of Meta's innovation leadership at a time when competition across the industry is intensifying from both established technology giants and agile startups such as Mistral AI. Analysts warn that this loss of talent could undermine Meta's previous advantage in developing open-source generative artificial intelligence. In particular, the growth of Mistral AI, which has managed to attract so many former Meta researchers, poses a direct challenge to Meta's market position. Mistral has quickly established itself as a major player in the European AI ecosystem, and its success is partly built on expertise acquired from former Meta employees. The situation also reveals a current trend in the technology industry, where top AI researchers are gaining ever-greater bargaining power and opportunities. The startup environment offers not only better financial terms but also greater flexibility and the opportunity to participate in shaping the future of artificial intelligence from the ground up.

Meta must now confront the challenge of replacing the lost talent and rebuilding its innovation capacity in the critically important field of AI research. Success in this area will be crucial for maintaining the company's competitiveness in the rapidly evolving artificial intelligence landscape.

Category:AI
Did you enjoy this article?
Discover more interesting posts on our blog
Back to blog

Related posts

Altman Announced the Singularity Days After His Models Escaped the Lab on Their OwnAltman Announced the Singularity Days After His Models Escaped the Lab on Their Own
OpenAI chief Sam Altman declared on the Relentless podcast that humanity has already entered the singularity. “We’re like, in the singularity now,” he said verbatim. For decades, the term belonged more to science-fiction literature
6 min read
28. 7. 2026
AI Remixed a Madonna Song—and Now It Tops the Charts in Australia. Musicians Are Furious.AI Remixed a Madonna Song—and Now It Tops the Charts in Australia. Musicians Are Furious.
Since April, Australian radio has been playing a dance remake of Madonna’s hit Like a Prayer on repeat. Released by Queensland DJ Josh Fawaz, it tops the radio airplay chart and has 35 million Spotify streams.
6 min read
28. 7. 2026
Claude Opus 5 Built a Shooter from Scratch. What Can Claude of Duty Do?Claude Opus 5 Built a Shooter from Scratch. What Can Claude of Duty Do?
A first-person shooter that runs directly in the browser, with its own physics and eleven separate code modules. Around 55,000 lines in total, split across eleven subsystems and built on Thr
4 min read
28. 7. 2026
Přihlaste se k odběru našeho newsletteru
Zůstaňte informováni o nejnovějších příspěvcích, exkluzivních nabídkách, a aktualizacích.
CodedTrip

Operated by CodedTrip LLC, USA.

YouTube
TikTok