Mark Zuckerberg, the chief executive of Meta, said in July that his company’s new artificial intelligence models would “push the frontier in the next year or so.”
Now Mr. Zuckerberg — who has invested billions in the A.I. race — appears increasingly unlikely to hit that deadline, three people with knowledge of the matter said.
Meta’s new “frontier” A.I. model, which the company has been working on for months, has fallen short of the performance of leading A.I. models from rivals like Google, OpenAI and Anthropic on internal tests for reasoning, coding and writing, said the people, who were not authorized to speak publicly about confidential matters.
The model, code-named Avocado, outperformed Meta’s previous A.I. model and did better than Google’s Gemini 2.5 model from March, two of the people said. But it has not performed as strongly as Gemini 3.0 from November, they said.
As a result, Meta has delayed Avocado’s release to at least May from this month, the people said. They added that the leaders of Meta’s A.I. division had instead discussed temporarily licensing Gemini to power the company’s A.I. products, though no decisions have been reached.
How Meta’s A.I. model performs is being closely watched in the competition over the fast-evolving technology. Google, OpenAI and Anthropic are widely regarded as the leaders in frontier A.I. models, which are the basis for developing new chatbots, video generators, coding tools and other products. Being at the forefront of A.I. development also helps companies recruit technologists and keep up a stream of experimentation.
Mr. Zuckerberg, 41, has staked the future of Meta, which owns Facebook, Instagram and Threads, on being at the cutting edge of A.I. His company has spent billions hiring top A.I. researchers and committed $600 billion to building data centers to power the technology. In January, Meta projected that it would spend as much as $135 billion this year, nearly twice the $72 billion it spent last year.
It takes time to improve frontier A.I. models, and Meta can still catch up to rivals, A.I. experts said. But a longer timeline has set in at the company, with Mr. Zuckerberg tempering expectations for Avocado in the past few months.
“I expect our first models will be good, but more importantly will show the rapid trajectory we’re on,” he said on a call with investors in January.
A spokesman for Meta, Dave Arnold, said in a statement on Thursday: “As we’ve said publicly, our next model will be good but, more importantly, show the rapid trajectory we’re on, and then we’ll steadily push the frontier over the course of the year as we continue to release new models. We’re excited for people to see what we’ve been cooking very soon.”
(The New York Times sued OpenAI and Microsoft in 2023, accusing them of copyright infringement of news content related to A.I. systems. The two companies have denied those claims.)
Mr. Zuckerberg bet big on a new A.I. model after Meta’s previous model, Llama 4, fell short of expectations last year. To prevent further setbacks, the company invested $14.3 billion in the start-up Scale AI in June and made its chief executive, Alexandr Wang, 29, its new chief A.I. officer. Mr. Zuckerberg declared that Meta’s new goal was to create a “superintelligent” form of A.I. that would lead to “a new era for humanity.”
Mr. Wang helped assemble an elite A.I. lab within Meta called TBD Lab (for “to be determined”), which began working on two new fruit-themed A.I. models — Avocado and Mango, an image and video generator.
TBD Lab finished the first stage of Avocado’s development, called “pre-training,” at the end of last year. In January, it began the next phase, “post-training,” which is when the team set a target release date of mid-March, two people with knowledge of the matter said.
So far, the new A.I. division has released one product — Vibes, an A.I. video app similar to OpenAI’s Sora.
Meta’s executives have debated whether the new A.I. model will be “open source,” which means parts of its code are public for other developers to build on, or closed so the underlying code remains private. Meta has long championed open source models, arguing that they help advance the technology, while companies like OpenAI and Anthropic have said letting others build off their A.I. would pose safety risks.
Over the summer, Mr. Zuckerberg and Mr. Wang leaned toward making Meta’s new model closed, two people with knowledge of the matter said.
TBD Lab, which has around 100 employees, has been hiring and has experienced some turnover, with a handful of researchers departing before Avocado’s release.
Mr. Wang has also clashed with Chris Cox, Meta’s chief product officer, and Andrew Bosworth, the chief technology officer, over how the new A.I. models should improve the company’s advertising business.
Last week, Meta said in a note to employees, which was reported earlier by The Wall Street Journal, that it would create an A.I. engineering team under Mr. Bosworth that would collaborate with Mr. Wang and the A.I. division.
Rumors soon swirled that Mr. Zuckerberg and Mr. Wang were on the outs. Meta moved quickly to squelch the talk, with a spokesman calling the idea “totally false.” On Threads on Monday, Mr. Zuckerberg posted a selfie of him and Mr. Wang with the caption “Meanwhile at Meta HQ.”
Meta’s leaders are already thinking big about future A.I. models. Its next one will be named after an even larger fruit, Watermelon.
Kalley Huang contributed reporting. Sheelagh McNeill contributed research.
Eli Tan covers the technology industry for The Times from San Francisco.
The post Meta Delays Rollout of New A.I. Model After Performance Concerns appeared first on New York Times.
