Meta Platforms has pushed back the launch of its next flagship artificial intelligence model, code‑named “Avocado,” to May or later, according to multiple media reports based on conversations with people familiar with the project. The delay highlights both the technical challenges Meta faces in closing the gap with leaders like OpenAI and Google, and the company’s reluctance to roll out a marquee model that does not clearly meet the latest performance expectations.
The New York Times reported that Meta has postponed Avocado’s rollout from an internal target this month to “at least May,” citing several people with knowledge of internal discussions. U.S. News & World Report, summarising a Reuters dispatch, noted that Meta has “pushed the rollout of its new AI model, code‑named ‘Avocado,’ to May or later,” moving beyond the previously expected March debut.
Coverage in The Economic Times said Meta now expects to introduce Avocado sometime “in May or June,” adding that the updated window underscores the company’s focus on getting performance right before a wide release. The model has been in development for months and is intended to be a central pillar of Meta’s AI strategy for 2026, following earlier signals that the company is working on next‑generation systems internally referred to as “Mango” and “Avocado.”
The decision to delay is tied primarily to how Avocado stacks up against rival systems, rather than any single technical failure. A report in The Business Times, relaying details from people briefed on internal testing, said Avocado outperformed Meta’s existing model and “did better than Google’s Gemini 2.5 model from March,” but “has not performed as strongly as Gemini 3.0 from November.”
Those tests focused on reasoning, coding and writing—capabilities that now define the top tier of large language models and heavily influence perceptions of which company is in front. People familiar with the matter told reporters that Meta’s leadership concluded Avocado was not yet competitive enough with the very latest models from rivals, particularly Google, to justify a high‑profile launch this month.
In a sign of how intense the AI race has become, senior figures at Meta have discussed using Google’s Gemini technology as a temporary underpinning for some AI features, according to the New York Times account. These talks, described as exploratory, would allow Meta to license a rival’s model to power parts of its products while Avocado continues to be refined, although there is no indication that a final decision has been taken.
Such a move would be unusual between two of Silicon Valley’s largest competitors, but it illustrates how quickly the technical bar is rising. Depending on an external model, even for a limited period, would underline the pressure on Meta to deliver a strong in‑house system and reduce any perception that it is trailing in core AI capabilities.
Publicly, Meta has stayed away from commenting directly on internal benchmark results or potential licensing discussions. Instead, the company has stressed that more advanced models are on the way and framed 2026 as a year of rapid AI progress.
In a statement to Reuters, quoted by outlets including Yahoo Finance, a Meta spokesperson said: “Our upcoming model will be impressive, but more importantly, it will demonstrate the rapid progress we are making, and we will consistently push the boundaries throughout the year as we continue to unveil new models.” The spokesperson added, “We’re eager for everyone to experience what we have been preparing very soon.”
That message is aimed at encouraging investors and users to see Avocado as part of a broader pipeline of models and upgrades rather than a single make‑or‑break release, with further systems expected after its launch.
The delay comes as Meta dramatically steps up its investment in AI infrastructure and research. Earlier this year, the company projected capital expenditure of at least 115 billion dollars, with much of that earmarked for AI chips, data centres and large‑scale model training. Chief executive Mark Zuckerberg has previously outlined a long‑term ambition to build systems moving towards “superintelligence,” and Meta has created a dedicated unit, Meta Superintelligence Labs, to drive that effort.
Reports published in late 2025 indicated that Meta is working on two major next‑generation models—Mango and Avocado. Mango is expected to focus on advanced image and video generation, while Avocado is aimed at higher‑end reasoning and coding tasks, positioning the company to compete more directly with OpenAI’s latest GPT models, Google’s Gemini family and new offerings from Anthropic.
Meta’s current Llama models, many of which are released under open licences, have been widely adopted by developers but are often regarded as sitting just behind the most powerful proprietary systems in complex reasoning and multi‑step problem solving. Avocado is intended to close that gap and act as the core model behind more capable Meta AI features across Facebook, Instagram, WhatsApp and other services.
According to earlier coverage of internal briefings, Meta has described Avocado as its “most ambitious LLM to date” and linked the work to richer “world models” that aim to understand and predict how real‑world systems behave, rather than only producing fluent text. That ambition helps explain why internal teams are wary of releasing a version that does not clearly match or beat top alternatives on headline benchmarks.
The updated Avocado timeline lands in a crowded AI landscape. Google has been rapidly extending its Gemini line across consumer and enterprise products, while OpenAI continues to update its GPT‑series models in partnership with Microsoft. Analysts have repeatedly said that Meta’s ability to justify its surging AI spend will depend not only on the timing of Avocado’s launch, but on how convincingly it performs against rivals in real‑world use and how deeply it is integrated into revenue‑generating tools, from advertising products to consumer assistants.
For now, Avocado is still expected to arrive later this spring or in early summer, after additional training and fine‑tuning. Whether or not Meta briefly leans on Google’s Gemini or other external technology along the way, the eventual rollout will be viewed as a key test of whether the company’s heavy AI bet is starting to translate into capabilities that users encounter every day inside Meta’s apps and services.
Comments