Meta Delays ‘Avocado’ AI Model as Performance Falls Behind Rivals

Title: Meta Reportedly Considers Licensing Google Gemini as Internal AI Efforts Stall

Meta's ambitious plans to compete at the frontier of artificial intelligence have hit a significant roadblock, with the company's latest large language model — internally code-named Avocado — being delayed until at least May. The setback comes nine months after CEO Mark Zuckerberg assembled what he called an AI "super team" and committed over $14 billion to the effort.

According to people familiar with the matter, Avocado outperforms Meta's previous AI models and surpasses Google's Gemini 2.5. However, it falls short of the newer Gemini 3.0, as well as the latest offerings from OpenAI and Anthropic. For a company that has staked its future on AI leadership, the gap is reportedly causing concern among senior leadership.

Perhaps most striking is the revelation that Meta's leadership has discussed the possibility of licensing Google's Gemini technology on a temporary basis — a move that would represent a remarkable acknowledgment of the competitive gap. Such an arrangement would be highly unusual in the technology industry, where companies typically view their AI capabilities as core competitive assets.

The delay underscores the difficulty of building frontier AI systems, even with virtually unlimited resources. Meta has been hiring aggressively in the AI space, including bringing on former Scale AI CEO Alexandr Wang. Despite these investments, the company continues to trail its primary competitors in benchmark performance.

Analysts have noted that Meta's situation reflects a broader trend in the AI industry, where the gap between leading and trailing companies appears to be widening rather than narrowing. The compute requirements for training state-of-the-art models have grown exponentially, and subtle differences in training data, architecture choices, and engineering culture can compound into significant performance gaps.