Meta reportedly delays rollout of new AI model Avocado – here's why

Meta's artificial intelligence model code-named “Avocado” will face another delay, the New York Times reported on Thursday. The release of this Meta AI model has been moved to at least May from March.

According to media reports in December, Meta was working on a text AI model code-named Avocado slated for a first-quarter launch.

The delayed timeline comes even as the company invests heavily to expand its AI ambitions, including a roadmap for building its own chips.

Here's why ‘Avocado’ is facing a delay:

Avocado, Meta's new AI model, has fallen short in performance when compared to the latest offerings of its rivals, the NYT report said.

The company has been working on it for months.

The leaders of Meta's AI division also explored the possibility of temporarily licensing Google's Gemini to power the company's AI products, the NYT report said. However, no decisions have been reached yet.

“Our next model will be good, but more importantly, show the rapid trajectory we're on, and then we'll steadily push the frontier over the course of the year as we continue to release new models,” a Meta spokesperson told Reuters.

"We're excited for people to see what we've been cooking very soon," the spokesperson added in an emailed statement.

About Avocado

Avocado is optimised for logical reasoning, software development, and "agentic" behaviour—meaning it is built to act as an autonomous agent that can plan and execute multi-step tasks rather than just generating text.

Internal reports suggest that Avocado is a significant leap over Llama 3.x, though recent leaks indicate it is currently performing somewhere between Google’s Gemini 2.5 and Gemini 3.

It is being developed alongside a second model codenamed "Mango," which focuses specifically on high-fidelity image and video generation to compete with tools like OpenAI's Sora, reported Business Times.

Meta's pursuit of ‘superintelligence’

In January, Meta laid out capital spending plans of between $115 billion and $135 billion for the year in the pursuit of "superintelligence" — the horizon where AI will outsmart humans.

“As we plan for the future, we will continue to invest very significantly in infrastructure to train leading models and deliver personal super intelligence to billions of people and businesses around the world,” Mark Zuckerberg was quoted as saying by CNBC during the earnings call.

The "infrastructure" Zuckerberg cited during the earnings call refers to Meta's ambitious data centre projects that will anchor both its current and future AI projects.

The Meta CEO also stressed that 2026 would be a major year for AI, with the company's investments aimed at advancing his mission for creating a “personal super intelligence,” which he announced last year.