The model had been expected to launch this month, but sources told Reuters that internal tests showed it was still lagging behind Google’s latest Gemini systems, leading Meta to push development into late spring.

People familiar with the project say Avocado currently performs somewhere between Gemini 2.5 and Gemini 3, leaving it short of the benchmark Meta had hoped to hit before unveiling the technology.

The company has been working on the model for months as part of its wider push to compete at the top end of the AI market.

The delay comes as Meta continues to pour vast resources into advanced AI research.

In January, the company outlined capital spending of $115–135 billion for the year, tied to its long‑term ambition to build what CEO Mark Zuckerberg has described as “superintelligence.”

Despite the setback, Meta insists the new model will demonstrate the rapid progress its AI division is making.

A spokesperson said the upcoming release will be “good” and will show the trajectory Meta is on, echoing comments Zuckerberg made during the company’s most recent earnings call.

They added that Meta plans to roll out further models throughout the year.

Executives inside Meta’s AI unit have also discussed the possibility of temporarily licensing Google’s Gemini to support some of the company’s AI products while Avocado continues development, though no decision has been made.

The postponement highlights the increasingly competitive race among major tech firms to deliver the most capable AI systems. Meta has been positioning itself as a key player in the next generation of AI models, but the delay underscores the challenge of keeping pace with rapid advancements from rivals.