Meta’s next AI model could mark a change in its “open source” AI strategy that Mark Zuckerberg had called “the path forward.” Currently, the company is working on a new AI model, code-named Avocado, that it might charge for access to, Bloomberg reports.
Last year, Meta launched its open source (depending on who you ask, since the Open Source Initiative disagrees) Llama 4 AI model, but it had a disappointing release, with Meta caught gaming AI benchmarks and being forced to delay a planned “Behemoth” version. But Zuckerberg scrapped that “in pursuit of something new,” Bloomberg says.
In a July 30th memo about “personal superintelligence,” Zuckerberg indicated that Meta may need to shift its approach on open source: to mitigate potential safety risks, the company will have to be “careful about what we choose to open source,” he said.
As part of the changes, Zuckerberg now “spends much of his time and energy working closely with those new hires, in a group called TBD Lab,” Bloomberg says. That team has a “siloed space” near Zuckerberg’s office at Meta’s headquarters, The New York Times says in its own report about the company’s AI efforts.
