MENLO PARK, Calif. — Luke Metz, a veteran AI researcher who spent years at OpenAI and briefly at the startup Thinking Machines, is joining Meta’s Superintelligence Labs, a person familiar with the matter told Axios. Metz is expected to start this week, reporting to Alexander Wang, the executive Meta placed in charge of all of its AI work. The hiring, confirmed by a person familiar with the arrangement, is the latest in a series of moves by Meta to stock its frontier-research teams with senior talent.
Metz’s résumé traces the recent history of the field’s most competitive lab. He joined OpenAI during its early research era, worked on reinforcement learning and agents, then left in 2024 to join Thinking Machines, the startup founded by Mira Murati after she departed OpenAI. Earlier this year he returned to OpenAI, and now he is leaving again — this time for Meta. The pattern of departures, returns and departures reflects a market where researchers hold unusual power, and where the biggest companies are willing to pay heavily for people who can build the next generation of models.
At Meta, Metz lands in a lab with an aggressive mandate. Superintelligence Labs, which the company has been staffing up over the past year, is tasked with pushing beyond the current generation of large language models toward systems with longer planning horizons and greater autonomy. Meta has said little publicly about the group’s charter, but its hiring has been unmistakable: the company has pulled in researchers from OpenAI, Google DeepMind and academic labs, and it has opened offices near rival research hubs to make recruiting easier.
Wang, Metz’s new boss, runs the company’s entire AI organization, spanning the research labs, the product teams that ship Llama models, and the infrastructure group that builds the data centers where those models train. Putting a frontier researcher under Wang rather than under the separate research division is a deliberate structure: Meta wants the researchers who build models close to the teams that deploy them. The company has been shipping new AI models at a rapid clip in recent months, and executives have said publicly that they intend to keep the pace, treating releases like a software cadence rather than occasional events.
The hire also fits Meta’s broader strategy of turning AI research into products that ship. The company’s Llama models are open-source, distributed free to developers in a bid to build an ecosystem that can challenge OpenAI’s closed models. Frontier researchers like Metz are prized not only for what they build but for the credibility they bring: when a company announces that its next model was built by the people who shaped the field, developers listen.
For OpenAI, the departure is a familiar kind of loss. The company has seen a steady stream of senior researchers leave over the past three years, some to found startups, some to join rivals. OpenAI has responded by promoting from within and by moving quickly to new model generations, but the churn has given its competitors a recruiting story of their own: that the talent that made OpenAI’s early breakthroughs is now dispersed across the industry, building similar capabilities elsewhere. Meta has been the most aggressive beneficiary, and its Superintelligence Labs has become the landing spot for several high-profile arrivals.
The broader pattern is a market for AI talent that behaves like a market for elite athletes. Compensation packages run into the eight figures, contracts include retention bonuses with staggered vesting, and poaching is open and constant. Analysts who follow the sector say the flow of researchers from OpenAI to Meta is notable because the two companies are now direct rivals in frontier models, with Meta’s Llama competing against OpenAI’s GPT and Claude lines in benchmarks and in enterprise deals alike.
Metz’s start date, this week, means he will join just as Meta begins its fall product cycle, a period when the company typically previews new models and features. People close to the company said Metz is expected to work on the lab’s agent-focused research, an area Meta has identified as central to its next generation of products. The arrival of a researcher who spent years on reinforcement learning and agent behavior at OpenAI suggests Meta intends to compete not just on model size but on what the models can do on their own. For a company whose AI ambitions have sometimes been described as trailing its rivals, the hire is a statement about where it wants to be in the next phase of the race.
Meta’s recruiting push is not limited to researchers. The company has also been hiring engineers and infrastructure specialists from rival labs, and it has expanded its AI compute budget to match its ambitions. Earlier this year Meta said it would spend tens of billions of dollars on data centers and chips, a figure that puts it in the same league as Microsoft and Google. The combination of capital and talent is the standard formula for frontier AI, and Meta’s ability to attract people like Metz suggests the formula is working.
The hire also carries a message to the broader research community about where Meta wants to stand. For years the company’s AI reputation was built on open-source releases and on the scale of its social platforms. The Superintelligence Labs project signals a bid for a different kind of standing: a lab that competes at the frontier, publishes at the top conferences, and builds systems that researchers describe as genuinely new. Metz, who has authored influential papers on learning algorithms and agents, fits that ambition. His arrival gives the lab a researcher whose name is known to every graduate student in the field, and whose work history spans both of the companies that defined the modern AI race.
