Meta Delays Next-Gen AI Model 'Avocado' to May+ as Performance Falls Short of Google and OpenAI

Meta Delays Next-Gen AI Model 'Avocado' to May+ as Performance Falls Short of Google and OpenAI

In March 2026, it was revealed that the release of Meta's next-generation AI model, codenamed "Avocado," has been postponed from its originally scheduled March 2026 launch to at least May or later. The primary reason for the delay is that internal testing showed the model fell short of key models from competitors Google, OpenAI, and Anthropic in the areas of logical reasoning, programming, and writing.

Specifically, while Avocado reportedly outperformed Meta's existing models and Google's "Gemini 2.5," it failed to match the latest "Gemini 3.0." In response to these results, Meta's leadership reportedly considered temporarily licensing Google's Gemini, though no final decision has been reached.

Meta has been investing billions of dollars in the AI field, including a $14.3 billion investment in AI data company Scale AI, whose CEO has been brought on to lead Meta's frontier AI division. Development of "Watermelon," the next-generation model following Avocado, is also underway, alongside "Mango," a model specializing in image and video generation.

Particularly noteworthy is the ongoing internal debate at Meta over whether to release its new AI models as open-source or keep them closed. The company has been known as a champion of open-source AI through its Llama series of open-weight models, but intensifying competition may be prompting a strategic reassessment.

This development delay vividly illustrates the fierce competition in the AI field and the difficulty even tech giants like Meta face in keeping pace with the rapid speed of technological innovation. The AI development race has entered a new phase, and each company's strategic decisions will significantly shape the future competitive landscape.