Meta’s delayed Avocado AI model has sparked an internal debate on whether to maintain its open source strategy or temporarily license Google’s Gemini as competition from OpenAI and Anthropic intensifies.
Meta Platforms is delaying the release of its next artificial intelligence model, Avocado, by about two months after internal testing showed it underperformed against rival systems, triggering an internal debate over whether the company should maintain its open-source AI strategy.
The model, initially expected to launch in March, is now likely to debut around May after tests showed it lagging behind competing models in key capabilities including reasoning, coding and writing. While Avocado performed better than Meta’s previous model and surpassed Gemini 2.5 released in March, it still trailed Gemini 3.0 released in November.
The disappointing results have prompted Meta leaders to consider temporarily licensing Google’s Gemini technology to power some of its products.
At the same time, senior executives are reportedly debating whether Avocado should follow the company’s traditional open-source model approach. Meta has long promoted open-source AI, arguing that making models available to developers accelerates innovation. Rival firms such as OpenAI and Anthropic, however, favour more controlled model releases due to safety concerns.
The situation highlights a broader industry divide between open and proprietary AI development strategies.
Meta has invested heavily to compete in the AI race, committing roughly $600 billion toward data-centre infrastructure and projecting up to $135 billion in AI spending this year, compared with $72 billion last year.
“I expect our first models will be good, but more importantly will show the rapid trajectory we’re on,” said Mark Zuckerberg, chief executive officer of Meta.
AI development is being led by Alexandr Wang, Meta’s chief AI officer and founder of Scale AI, who set up an internal unit called TBD Lab to build the Avocado model. The roughly 100-person lab completed the model’s pre-training stage late last year and began post-training in January.
The division has so far launched Vibes, an AI video-generation app similar to Sora.
Internal tensions have also emerged over AI strategy, with reported disagreements between Wang, chief product officer Chris Cox and chief technology officer Andrew Bosworth regarding how AI models should strengthen Meta’s advertising business. The company recently formed a new AI team under Bosworth to work alongside Wang’s group.
A Meta spokesperson said: “We’ll steadily push the frontier over the course of the year as we continue to release new models. We’re excited for people to see what we’ve been cooking very soon.”
