Meta has brought in Trapit Bansal, a highly influential OpenAI researcher, to work on AI reasoning models for the company’s new AI superintelligence unit, according to a person familiar with the matter cited by TechCrunch.
OpenAI spokesperson Kayla Wood confirmed to TechCrunch that Bansal had left OpenAI. Bansal’s LinkedIn page says he departed in June.
Why Trapit Bansal matters
Bansal has been at OpenAI since 2022, and his role there connects directly to the type of AI capability Meta is now trying to strengthen. He was a key player in starting OpenAI’s reinforcement learning work alongside co-founder Ilya Sutskever.
He is also listed as a foundational contributor on OpenAI’s first AI reasoning model, o1. That background gives Meta a researcher with direct experience in one of the areas that has become central to the latest phase of AI competition.
For Meta, the hire is not simply about adding another senior researcher. It is about bringing in someone whose work overlaps with the exact technical gap Meta is trying to close: frontier AI reasoning.
Meta’s superintelligence team is taking shape
Bansal is joining Meta’s new AI superintelligence lab, a group that already includes leaders such as former Scale AI CEO Alexandr Wang. Meta is also looking to add former GitHub CEO Nat Friedman and Safe Superintelligence co-founder Daniel Gross.
Other high-profile AI researchers have recently joined the same effort. The Wall Street Journal reported on Wednesday that Lucas Beyer, Alexander Kolesnikov, and Xiaohua Zhai, all former OpenAI researchers, have joined Meta’s AI superintelligenc e team in recent weeks.
Bloomberg reported that Bansal will join them alongside former Google DeepMind researcher Jack Rae and Johan Schalkwyk, a former machine learning leader at the startup Sesame.
The pattern is clear from the source: Meta is building this group through targeted hiring of people with deep experience at leading AI organizations. Mark Zuckerberg has also been trying to grow the team through other routes.
Zuckerberg’s AI hiring campaign
In recent months, Zuckerberg has been on a hiring spree to build Meta’s new AI team. TechCrunch reports that Meta has offered $100 million compensation packages to top researchers who join the company. It is unclear what Bansal was offered in this deal.
Zuckerberg reportedly also tried to acquire startups with major AI research labs, including Sutskever’s Safe Superintelligence, Mira Murati’s Thinking Machines Labs, and Perplexity. Those talks did not reach a final stage.
The pressure around AI talent has also become public. On a recent podcast, OpenAI CEO Sam Altman said Meta had been trying to recruit top people from his startup, but claimed that "none of our best people have decided to take him up on that.”
A Meta spokesperson declined to comment.
Why AI reasoning is the strategic target
AI reasoning models are a key area for Meta’s new team because they have become one of the clearest ways leading AI labs are pushing model capability forward. In the last year, OpenAI, Google, and DeepSeek have released highly performant AI reasoning models.
The basic idea, as described in the source, is that these systems are trained to work through problems before producing an answer. They use additional time and computing resources in that process. AI labs have found that approach can improve performance on benchmarks and real-world tasks.
Meta currently does not offer a public AI reasoning model. That makes Bansal’s arrival especially important for the company’s competitive position.
TechCrunch notes that Bansal could help Meta develop a frontier AI reasoning model able to compete with industry-leading technology such as OpenAI’s o3 or DeepSeek’s R1. For Meta, that would be more than a research milestone. It could feed directly into products across the company.
How reasoning models could support Meta’s products
Meta’s AI superintelligence lab could become an internal engine for products throughout the company, similar to the role Google’s DeepMind unit plays. The source points in particular to Meta’s effort to build AI agents for business under former Salesforce CEO of AI Clara Shih.
Those agents would need strong reasoning models behind them if Meta wants them to be competitive. A business AI agent is not just expected to generate text; it must handle tasks, evaluate steps, and produce useful answers in situations where the right response may require more than a quick prediction.
That is why the reasoning model push matters. It sits behind the more visible product goals.
With Bansal and other researchers now joining, Meta is trying to pull ahead in the AI race. But the challenge may soon become harder: OpenAI plans to release an open AI reasoning model in the coming weeks, a move that could increase pressure on Meta’s open AI offerings.