It has been months since Mark Zuckerberg’s Meta has been pushing to get ahead in the AI race. The Meta Superintelligence Labs (MSL), led by former Scale AI founder Alexandr Wang, has been collecting some of the brightest minds in the AI space to develop a frontier AI model and beat rivals like Google, Anthropic, and OpenAI. Sadly, though, a report now suggests that Meta AI’s original plan to launch its next-generation AI model may not be on track, as the company has delayed the release from March 2026 to May or June 2026.
The AI model, internally known as Avocado, is expected to compete with the most advanced AI models from rivals like OpenAI, Google, and Anthropic. A report from The New York Times now suggests that Avocado’s release has been delayed until May or June 2026, owing to performance issues.
Meta AI delays release of next-gen Avocado model
The delay emerges from the model’s current performance falling short of leading competitors in key benchmarks, including reasoning, coding, and writing capabilities. Sources familiar with the matter stated that Avocado – a model that Meta has been developing for months as a text-focused large language model (LLM) – outperforms the company’s previous open-source Llama models.
However, it lags behind top offerings from rivals like Google’s Gemini series. More specifically, Avocado’s capabilities currently sit between Google’s Gemini 2.5 and the more advanced Gemini 3, prompting internal leaders to hold off on release until further performance improvements can be made.
While the delay seems like a negative aspect for Meta at a time when rivals are pushing to compete at the highest levels of AI model development, Meta seems committed to releasing an AI model that offers meaningful progress over marketing efforts.
Meta’s response to the delay
A Meta spokesperson commented on the delay of the AI model’s release in a statement to Reuters, stating, “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. We’re excited for people to see what we’ve been cooking very soon, ” stated Meta.
The delay comes at a time when rumours of a partnership with Google Gemini emerge. Industry insiders suggested a temporary licensing of Google’s Gemini AI models to power Meta’s AI products until the company manages to cook something competitive. However, those rumours remain unverified.
The Avocado model is said to be part of Meta’s broader 2026 AI roadmap, which also includes a multimodal model code-named “Mango” focused on image and video generation. The Facebook parent company aims to release a series of continuously improving AI models throughout the year to maintain momentum in the generative AI space.