Meta delays flagship AI model Avocado to May 2026 at the earliest

Meta has delayed the release of its new flagship AI model Avocado: instead of March 2026, the launch is now expected no earlier than May. According to the NYT and Reuters, the company decided to pause after internal evaluations in which the model fell short of the best systems from competitors.

Why the release was pushed back

Western outlets report that the problem is not that Avocado is weak in itself. In internal tests it beat Meta's previous models, but it did not deliver the leap the company had expected. Reuters' sources say the model's answer quality currently sits between Google Gemini 2.5 and Gemini 3.0. For a project meant to showcase a new phase of Meta's AI strategy, that level was not enough. So the company chose not to force the release.

Instead of a March launch, Meta has reportedly moved the release window to May at the earliest, and according to some reports even to June. The delay shows how brutal the frontier-model race has become: it is no longer enough to ship "yet another strong model." You either have to reach the level of the leaders or offer a very clear advantage in price, speed or product integration.

What we know so far

The published details point not to a failure but to a drawn-out polishing phase. Meta is not shelving the project and, on the contrary, keeps pouring huge resources into AI, from compute infrastructure to its own chips. In January the company said its annual capital expenditures could reach $115 billion to $135 billion. Against that backdrop, the Avocado delay looks especially sensitive: the bigger the budget and expectations, the more noticeable every schedule slip.

• The model was originally planned for release in March 2026.
• The new target is May 2026 or later.
• According to media reports, Avocado is already better than Meta's earlier models but trails the latest Gemini.
• Leadership even discussed temporarily licensing Gemini for Meta's AI products.

The possible temporary Gemini license stands out separately. There is reportedly no final decision on that option yet, but the discussion itself is telling. Meta has bet on its own AI ecosystem for years, yet here it allows for a scenario in which someone else's model temporarily fills a product gap. That points to a more pragmatic approach: if the in-house flagship is not ready, the business cares less about ideology than about how fast it can ship features to users.

Pressure on Meta

Avocado matters as more than just another large language model. For Meta it is a test of a new phase of its strategy, in which the company is trying to close the gap with Google, OpenAI and Anthropic not through one-off demos but through steady product releases. If a flagship has to be delayed because it does not look convincing enough next to competitors, that hits two things at once: the team's internal tempo and the outside perception of Meta's entire AI lineup.

"The next model will be good, but what matters more is that it will

show our trajectory."

This remark from a company representative reflects the current mode: Meta is trying to win not a single release but a long series of updates over the year. But the market judges the nearest launch. If Avocado arrives later and without a clear edge, competitors get extra time to strengthen their models in search, assistants, coding and enterprise use cases. For Meta, that means even a temporary pause turns into more than a delay: it becomes a strategic risk.

What it means

The Avocado delay shows a simple thing: even a company with a multibillion-dollar AI budget can no longer afford to ship a flagship on a wing and a prayer. Google, OpenAI and Anthropic have set the bar, and every new release is now compared not with the previous generation but with the best models on the market right now.

*Meta is designated an extremist organization and banned in Russia.