Mark Zuckerberg, CEO of Meta, he said in July that his company’s new AI models “will push the envelope in the next or so.”
It now appears that Mr Zuckerberg – who has invested billions in the artificial intelligence venture – will miss that deadline, three people with knowledge of the matter said.
Meta’s new “frontier” artificial intelligence model, which the company has been working on for months, lags behind the performance of leading AI models from competitors such as Google, OpenAI and Anthropic in internal tests of reasoning, coding and writing, said the people, who were not authorized to speak publicly on confidential matters.
The model, codenamed Avocado, outperformed the previous Meta AI model and outperformed Google’s Gemini 2.5 in March, the two people said. But it didn’t perform as strongly as November’s Gemini 3.0, they said.
As a result, Meta has delayed the Avocado release until at least May starting this month, the people said. They added that Meta’s AI division leaders have instead discussed temporarily licensing Gemini to power the company’s AI products, though no decision has been reached.
The performance of the Meta artificial intelligence model is closely watched in the competition for the rapidly evolving technology. Google, OpenAI, and Anthropic are widely regarded as leaders in the frontier models of artificial intelligence that underpin the development of new chatbots, video generators, coding tools, and other products. Being at the forefront of AI development also helps companies acquire technologists and maintain a flow of experimentation.
Mr Zuckerberg, 41, has staked the future of Meta, which owns Facebook, Instagram and Threads, on being on the cutting edge of AI. His company has spent billions hiring top AI researchers and devoted $600 billion to building data centers to power the technology. In January, Meta projected it would spend up to $135 billion this year, nearly double the $72 billion it spent last year.
Frontier AI models take time to improve, and Meta may still catch up to rivals, AI experts said. But there has been a longer timeline at the company, with Mr. Zuckerberg tempering expectations about avocados in the past few months.
“I expect our first models to be good, but more importantly, they’ll show the rapid trajectory we’re on,” he said on a call with investors in January.
Meta spokesman Dave Arnold said in a statement Thursday: “As we’ve said publicly, our next model will be good, but more importantly, it will show the rapid trajectory we’re on, and then we’ll keep pushing the envelope throughout the year as we continue to release new models. We’re excited for people to see what we’ve cooked up soon.”
(The New York Times sued OpenAI and Microsoft in 2023, accusing them of copyright infringement for news content related to AI systems. Both companies denied the claims.)
Mr Zuckerberg bet on the new AI 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 CEO Alexander Wang, 29, the new head of AI. Mr. Zuckerberg he declared that Meta’s new goal was to create a “superintelligent” form of AI that would usher in a “new era of humanity”.
Mr. Wang helped set up an elite artificial intelligence lab within Meta called TBD Lab (for “determination”), which began working on two new fruit-themed AI models – Avocado and Mango, an image and video generator.
TBD Lab completed the first phase of Avocado development, called “pre-training,” late last year. The next phase, “post-training,” began in January, which is when the team set a target release date of mid-March, two people with knowledge of the matter said.
The new AI division has released one product so far – Vibes, an AI video application similar to OpenAI Sora.
Meta’s leadership has been debating whether the new AI model will be “open source,” meaning that parts of its code are public for other developers to build on, or whether it will be closed, so the underlying code remains private. Meta has long championed open-source models, saying they help advance technology, while companies like OpenAI and Anthropic have said letting others build on their AI would pose a security risk.
Over the summer, Mr. Zuckerberg and Mr. Wang leaned toward closing the new Meta model, two people with knowledge of the matter said.
TBD Lab, which has around 100 employees, is hiring and has seen some turnover, with several researchers leaving before Avocado’s release.
Mr. Wang also clashed with Chris Cox, Meta’s chief product officer, and Andrew Bosworth, chief technology officer, on how the new AI models should improve the company’s advertising business.
Last week, Meta said in a memo to staff that it was reported earlier by The Wall Street Journal that he would create a team of AI engineers under Mr. Bosworth to work with Mr. Wang and the AI division.
Rumors soon swirled that Mr. Zuckerberg and Mr. Wang were on the way out. The meta quickly moved to muffle the speech, with the speaker calling the idea “absolutely bogus.” On Monday on Threads, Mr Zuckerberg posted a selfie of himself and Mr Wang with the caption “Meanwhile at Meta HQ”.
Meta executives are already thinking a lot about future AI models. His next one will be named after an even bigger fruit, watermelon.
Kalley Huang contributed reporting. Sheelagh McNeill contributed research.