Meta Delays Release of ‘Avocado’ AI Model After Tests Show it Trailing Rivals
The model was initially expected to launch this month but has now been pushed back to at least May.
Topics
News
- Amazon to Buy Globalstar in $11.6 Billion Deal to Expand Satellite Ambitions
- Nvidia Launches AI Models to Tackle Quantum Errors
- Novo Nordisk Signs OpenAI Deal to Bring Medicines to Patients Faster
- Novartis CEO Narasimhan Joins Anthropic Board
- Meta Expands Broadcom Alliance to Build AI Chips
- Rockstar Games Hit by New Cyberattack With Limited Data Accessed
Meta has delayed the release of its new artificial intelligence model, code-named Avocado, after internal tests showed the system trailing rival models from Google, OpenAI and Anthropic.
The model was initially expected to launch this month but has now been pushed back to at least May, The New York Times reported, citing people familiar with the matter.
Some internal benchmarks showed Avocado performing between Google’s Gemini 2.5 and Gemini 3 models, prompting Meta to extend development timelines.
Avocado has been under development for several months and is designed as a foundational AI model capable of supporting tools such as chatbots, coding assistants and other AI applications.
While internal evaluations showed the model outperforming Meta’s earlier systems and surpassing Google’s Gemini 2.5 in some tests, it has not matched the capabilities of newer models such as Gemini 3.
The delay comes as Meta continues to pour large amounts of funding into AI infrastructure and research.
In January, the company projected annual capital spending between $115 billion and $135 billion as part of its broader push toward building advanced AI systems.
Chief executive Mark Zuckerberg has positioned AI development as a key priority, describing the effort as part of a long-term pursuit of what he calls “superintelligence,” a future stage of AI development where machines could outperform humans in many cognitive tasks.
Meta has been investing heavily in computing infrastructure, talent and acquisitions to strengthen its position in the global AI race. The company has hired leading researchers, expanded data center capacity and made large investments in AI-focused firms.
Zuckerberg has also signaled that the first versions of the company’s next-generation models may not immediately lead the industry but would demonstrate rapid progress.
“I expect our first models will be good, but more importantly will show the rapid trajectory we’re on,” Zuckerberg said during a January earnings call.
A Meta spokesperson reiterated that message in a statement. “As we’ve said publicly, our next model will be good but, more importantly, show the rapid trajectory we’re on, and then we’ll steadily push the frontier over the course of the year as we continue to release new models,” the spokesperson said. “We’re excited for people to see what we’ve been cooking very soon.”
The delay has also sparked internal discussions about whether Meta should temporarily license competing technology to power some of its AI products.
Avocado is being developed inside a specialized research group within Meta focused on building next-generation AI systems. The division has also been working on another project code-named Mango, which is aimed at image and video generation.
Meta’s AI strategy has expanded rapidly over the past year. The company increased capital spending sharply to support the development of new data centers and computing infrastructure designed for large-scale AI workloads.
The performance benchmarks for models like Avocado are closely watched because foundational AI systems serve as the underlying engines for many emerging technologies, including conversational assistants, generative media tools and automated coding platforms.
Meta’s earlier model, Llama 4, faced criticism last year for falling behind competing systems, increasing pressure on the company to demonstrate progress in its next release.
