Overview
- Multiple outlets report Meta is developing a proprietary model called Avocado that the company could charge for rather than release as open source.
- Sources say the training pipeline distills from third‑party models including Google’s Gemma, OpenAI’s gpt‑oss, and Alibaba’s Qwen, raising technical and geopolitical considerations.
- The project is led within Meta’s TBD Lab by Chief AI Officer Alexandr Wang, who joined after a $14.3 billion Scale AI deal and is described as favoring closed models.
- Meta is reallocating resources toward this effort, with reports that Reality Labs was asked to cut about $2 billion from its budget to fund AI development.
- The pivot follows dissatisfaction with Llama 4 and the scrapping of an internal successor known as Behemoth, alongside recent layoffs in AI units and the departure of Yann LeCun; timing and details of Avocado remain unconfirmed with a reported early‑2026 window.