Meta has pushed back the release of its next flagship AI model, internally codenamed Avocado, from this month to at least May 2026, according to a New York Times report citing three people with knowledge of the matter.

Performance Falls Short

Internal benchmarks show Avocado trails Google's Gemini 3.0 — released last November — on key tasks including reasoning, coding, and writing. The model does outperform Meta's previous generation and the older Gemini 2.5 from March, but the gap with frontier models from Google, OpenAI, and Anthropic hasn't closed enough to justify a release.

The delay puts pressure on Meta CEO Mark Zuckerberg, who pledged in July 2025 that new models would "push the frontier in the next year or so."

A Licensing Stopgap?

Leaders inside Meta's AI division have reportedly discussed temporarily licensing Google's Gemini to power Meta's AI products while Avocado development continues — though no decisions have been reached. That would be a striking move for a company that has publicly championed open-source AI.

What's at Stake

Meta has invested billions in the AI race, betting that in-house foundational models are critical for recruiting talent and building competitive products. Avocado was supposed to be its strongest model yet and the first major release since Alexandr Wang, former CEO of Scale AI, joined to lead a revamp of Meta's AI efforts.

The setback highlights the difficulty of catching up in an AI landscape that Google, OpenAI, and Anthropic continue to push forward rapidly.