Meta Platforms has postponed the release of its new A.I. model, code-named Avocado, after internal tests showed it fell short of leading competitors. The company now expects to launch the model in May, rather than this month.

Meta’s chief executive, Mark Zuckerberg, has invested billions into A.I. research and development. The company’s previous model, Llama 4, underperformed expectations, prompting Meta to ramp up investment in its new A.I. division.

Avocado outperformed Meta’s older models and did better than Google’s Gemini 2.5, but lagged behind Gemini 3.0 from last November. Meta is reportedly considering temporarily licensing Gemini to power some products while Avocado is refined.

The model is being developed by TBD Lab, a team of around 100 employees led by Meta’s chief A.I. officer, Alexandr Wang. The lab has been working on two fruit-themed A.I. models — Avocado for text and reasoning, and Mango for image and video generation.

Meta executives are debating whether Avocado will be open source, which allows developers to build on the model, or closed, keeping the code private. The decision could affect adoption and safety concerns in the broader A.I. community.

So far, TBD Lab has released only one product, the A.I. video app Vibes, similar to OpenAI’s Sora. Meta plans to continue releasing updates gradually, showing progress over time rather than aiming for a single breakthrough product.

Despite the delay, Meta remains committed to its A.I. ambitions. Zuckerberg and Wang have outlined plans for future models, including one named Watermelon, aiming for “superintelligent” A.I. capabilities that could shape the company’s next era.