Meta Delays New AI Model. Its Loss Could Be Google’s Gain

Where is the next big artificial-intelligence model from Meta Platforms, Inc.? The social-media company is struggling to develop a bot that can match its peers and might even have to license Google technology, according to a report.

Meta hasn’t launched a cutting-edge AI model since its Llama 3 release back in 2024. That’s an eternity in the fast-moving industry and the company has seen the capabilities of its technology outstripped by Google’s Gemini, OpenAI’s GPT and Anthropic’s Claude since then.

Meta did release two versions of its Llama 4 AI model last year, but they were smaller models designed to run efficiently on limited hardware and were meant to be followed by a larger edition called “Behemoth.” But there has been no news since then.

The silence could be because in-house testing showed Meta’s latest AI model—now nicknamed “Avocado”— failed to perform as well as rivals and the company has decided to delay its release until at least May, the New York Times reported, citing sources. In the meantime, the Facebook parent is considering temporarily licensing Google’s Gemini to power its products, according to the NYT.

Such a move would be an embarrassment for Meta CEO Mark Zuckerberg, who has spent billions assembling a team of developers for his Superintelligence Labs division, which aims to develop the next generation of AI models, and has guided for up to $135 billion of capital expenditure this year, largely on computing infrastructure. And it would be more recognition for Google-parent Alphabet of its own transformation from perceived AI loser to winner.

Meta didn’t immediately respond to a request for comment from Barron’s early on Friday. A spokesperson told the NYT that the company was confident it would “push the frontier” in AI models over the course of the year.

Meta shares were down 3% in Friday trading.

Having the top AI model isn’t everything of course. Meta has already delivered returns on its AI investment in improvements in advertising and engagement, while also developing in-house chips to run its recommendation models efficiently and amassing a dominant share in the smart-glasses market.

However, Zuckerberg has very publicly set his sights on creating a Meta-developed “personal superintelligence,” with the eventual goal of his company’s combination of hardware and software supplanting Apple and other smartphone makers as the favored everyday computing device. At some point, that will require Meta to deliver a leading AI model.