Meta has hired highly influential Openai researcher Trapit Bansal to work on AI inference models under the company’s new AI Superintelligence unit.
Openai spokesman Kayla Wood confirmed to TechCrunch that Bansal has left Openai. According to Bansal’s LinkedIn page, he said he left Openai in June.
Bansal has been working at Openai since 2022 and, together with co-founder Ilya Sutskever, was a key player kickstarting the company’s work on reinforcement learning. He is listed as a fundamental contributor to Openai’s first AI inference model, O1.
Bansal can offer a big lift for Meta’s AI Superintelligence Lab, featuring leaders such as former scale AI CEO Alexandr Wang and also featuring leaders such as former Github CEO Nat Friedman and Safe Superintelligence co-founder Daniel Gross. Bansal helps develop frontier AI inference models that compete with industry-leading technologies such as Openai’s O3 and Deepseek’s R1. Currently, META does not publish AI inference models.
Over the past few months, Mark Zuckerberg has hired employment to build Meta’s new AI team, offering a $100 million reward package to top researchers joining his company. It is unclear what kind of Bansal was offered to participate in this agreement.
Despite this, Zuckerberg appears to have managed to acquire top AI research talent.
Three other former Openai researchers, Lucas Beyer, Alexander Kolesnikov and Xiaohua Zhai, have also joined Meta’s AI Superintelligence team in recent weeks, the Wall Street Journal reported Wednesday. According to Bloomberg, Bansal will join with former Google Deep Mind researcher Jack Ray and former machine learning leader Johann Sharquayk.
To fill out the new AI unit further, Zuckerberg reportedly tried to get a startup with a heavy hit AI research lab, including Sutskever’s safe super intelligence, Mira Murati’s Thinking Machines Labs, and confusion. However, these lectures did not proceed to the final stage.
In a recent podcast, Openai CEO Sam Altman argued that Meta is trying to poach top startup talents, but “none of our best people have decided to bring him to it.”
A Meta spokesman declined to comment.
The AI Inference Model presents the key areas for Meta’s AI Superintelligence team to get right. Last year, Openai, Google, and Deepseek shipped highly performant AI inference models that pushed the limits of what software can do. By training AI models, solving problems before answering, AI Labs has successfully improved the performance of AI in benchmarks and real-world tasks.
Meta’s AI Superintelligence Lab, like Google’s DeepMind unit, could become a key internal group that strengthens its products within the company. Meta is taking an ambitious effort to build AI agents for the business under Clara Shih, former Salesforce CEO of AI. To build a competitive agent, META needs to develop a reasoning model for frontier AI.
With the addition of Bansal and other leading AI researchers, Meta hopes to advance in the AI race. That can be difficult considering Openai plans to release an open AI inference model in the coming weeks.
This story will be updated after its release and details are provided.
