In the AI race, even the biggest players sometimes have to slam the brakes. For Meta, that moment arrived this week.
It’s reportedly paused the rollout of its next-generation AI model, nicknamed “Avocado,” until at least May, having previously aimed for an earlier release, after a series of internal tests showed the model still wasn’t quite meeting Meta’s standards. First reported here, it’s said to be ranked between two versions of Google’s Gemini AI in terms of its performance.
If you’ve been following AI developments, this pause isn’t so surprising. It’s not all about scaling up computing power; these AI systems are being tested for their reasoning powers, run over training data to see how they perform, and checked for how they operate in the real world. Further reporting here suggests that Avocado simply didn’t hit the performance thresholds it needed to match the best models being built by Google and OpenAI.
Let’s be clear: right now, Meta is under immense pressure. It’s pouring astronomical sums of money into AI research in a bid to catch up with the leaders in the field, including a planned capital expenditure this year of between $115-135 billion to build data centers and build AI infrastructure. This is part of a broader bid to pursue what Meta CEO Mark Zuckerberg has described as “superintelligence,” an AI system that can beat a human in most tasks.
Here’s where this gets awkward, and a bit intriguing for those of us who follow the tech sector: apparently, Meta execs have even considered temporarily licensing Google’s Gemini AI to use for some of Meta’s AI services while it waits for Avocado to get over the line. A bit like asking to borrow someone else’s engine until yours is fixed.
But before we get too critical, maybe that delay isn’t such a bad thing. The reality is that today’s AI systems are complex; they’re not just chatbots but AI being built to reason, to plan, to write software and act as autonomous agents. Avocado is expected to be capable of some of these advanced tasks, notably logical reasoning and multi-step problem solving.
And arguably, if Meta is taking a few more months before releasing an advanced AI system, that’s no bad thing. Nobody wants to release a model that wows in a presentation but breaks in the wild.
But for now, the AI race isn’t waiting. Google, OpenAI, Anthropic and a host of startups are pumping out new models at a frenetic pace. So Meta can’t keep Avocado on ice for long.
The real question now is: when it does get released, will it be enough to catch up, or even overtake?
Because in AI, today’s delay can be tomorrow’s breakthrough. Or tomorrow’s embarrassment. And right now, everyone in the tech sector is watching to see which one it will be.