Meta has effectively revised its open-source AI strategy and is said to be developing a closed model aimed at securing next-generation superintelligence. Observers say the company is moving in earnest to shift strategy after Llama4 fell short of market expectations.
On the 9th (local time), U.S. CNBC reported that Meta is developing the successor to the open-sourced "Llama" under the code name "Avocado" and plans to shift to a closed model that does not disclose core software elements such as weights. The initial target was to launch by the end of this year, but after internal testing the release has been adjusted to the first quarter of next year. Meta said, "Model training is proceeding as planned, and there are no major changes to the schedule."
Unlike OpenAI, Anthropic and Google, Meta had pushed an open strategy to quickly seize the market as a latecomer, but appears to have changed course after Llama4, unveiled in April, drew underwhelming reviews. CEO Mark Zuckerberg also stressed last year that "open source is the future," but this year said the company "must be judicious about what we share," shifting its stance. Internal backlash over China's DeepSeek R1 allegedly using Llama's design also played a role.
After Llama4, Zuckerberg pivoted the strategy to superintelligence development, investing $14.3 billion in Scale AI and hiring founder Alexander Wang as chief AI officer (CAIO). Avocado is being developed at Meta Superintelligence Lab (MSL) by the elite team "TBD Lab," led directly by Wang as CAIO. The team is said to operate virtually like a startup, separated from the network and located near the CEO's office.
Inside the company, pressure has grown to the point that 70-hour workweeks have become routine, and layoffs occurred during the reorganization. In the process, personnel moves continued, including Professor Yann LeCun, who had served as Meta's chief AI scientist since 2013, leaving to start a company.
This year Meta raised the lower end of its capital expenditure outlook to $70 billion from $66 billion and in October also moved to issue $25 billion in bonds, continuing large-scale investment. KeyBanc Capital Markets analyzed, "At the start of the year Meta was viewed as an AI winner, but it now faces questions about the scale of investment and profitability."