Meta has postponed the launch of its new base AI model, Avocado. Internal sources indicate that its performance in reasoning, programming, and writing tests did not reach the expected levels, falling below current benchmarks. This delay affects the entire chain of future products that depend on this foundational technology.
A fundamental model that did not meet the benchmarks 🧪
Avocado is designed as a foundational model, the base on which Meta planned to build chatbots and productivity tools. The problem is that, in comparative evaluations, it failed to surpass the benchmarks set by models from OpenAI and Google. This gap in key capabilities forced a technical review, prioritizing quality over the initial launch schedule.
An avocado that is not yet ripe? 🥑
It seems that the name Avocado was more prophetic than Meta expected. Like the fruit, which can go from hard to overripe in an instant, the model needs a very specific point of technical maturity. Perhaps the development team is looking for the perfect window: not too green to be useful, nor so soft that it generates hallucinations at the first complicated question.