Menlo Park, January 22, 2026 - Meta CTO Andrew Bosworth confirmed that the company’s Superintelligence Labs has delivered its first internal models just months after formation. The releases include a text-focused system codenamed Avocado and a multimodal image and video model codenamed Mango.

According to Bosworth, internal benchmarking shows strong performance across key tasks, though details about the architecture and training data remain undisclosed. The announcement is the first concrete milestone for the lab, which was formed to consolidate Meta’s frontier AI research efforts.

Industry analysts view the update as a signal that Meta is accelerating its race with other large labs. The internal delivery does not imply public release, but it suggests the company is moving toward production-grade systems at an aggressive pace.

Credit: Meta Superintelligence Labs team. Primary source: Reuters report.