Meta (Nasdaq: META) has pushed back the launch of its flagship AI model, codenamed Avocado, from March to at least May after internal tests showed it falling short of leading systems from Google (Nasdaq: GOOG), OpenAI, and Anthropic on logical reasoning, programming, and writing — a setback that has prompted the company’s leadership to discuss temporarily licensing Google’s Gemini technology as a stopgap.
The New York Times reported last week, citing three people with knowledge of the matter, that Avocado’s performance currently sits between Google’s Gemini 2.5 and Gemini 3.0, clearing the older model but falling short of the November 2025 release that Meta had hoped to match or beat.
The delay is the third major stumble in Meta’s frontier AI push. Llama 4 launched in April 2025 to a tepid reception, falling short of Zuckerberg’s January 2025 prediction that it would “become the leading state-of-the-art model” that year.
Behemoth — the flagship model in the Llama 4 series — stalled in May when engineers could not produce meaningful improvements.
Zuckerberg reorganized the entire effort under Meta Superintelligence Labs by June 2025, bringing in Scale AI founder Alexandr Wang as part of a $14.3 billion investment in Wang’s company, billing the initiative as the company’s path to personal superintelligence.
In January 2026, Zuckerberg told investors the first new models would show a “rapid trajectory,” a promise that now appears hard to keep.
Meta built its AI brand on the Llama series of freely available models, selling openness as the answer to closed ecosystems at OpenAI and Google. Borrowing Gemini, a rival’s model — even as a stopgap — guts that argument. No final decision has been confirmed.
Meta plans $115 billion to $135 billion in capital expenditures for 2026 in pursuit of what Zuckerberg calls superintelligence. Unlike Microsoft, Google, and Amazon (Nasdaq: AMZN), it has no cloud business to monetize what it builds, which makes Avocado’s commercial performance critical to justifying the spend.
The cost pressure is already landing on staff: Meta is planning layoffs affecting 20% or more of its roughly 79,000 employees to offset AI infrastructure costs in what would be its deepest cuts since the 2022–23 “year of efficiency.”
Meta spokesperson Andy Stone has called the news “speculative reporting about theoretical approaches.”
Information for this story was found via the sources and companies mentioned. The author has no securities or affiliations related to the organizations discussed. Not a recommendation to buy or sell. Always do additional research and consult a professional before purchasing a security. The author holds no licenses.
