Meta’s AI chief Alexandr Wang teases the company’s most capable model yet

Alexandr Wang, Meta Platforms’ Chief AI Officer, announced that the company will soon release its most capable AI model to date. The declaration signals an escalation in Meta’s AI ambitions under Wang’s leadership, positioning the social media giant for a more direct confrontation with OpenAI, Anthropic, and Google in the frontier model race.

Wang framed the company’s recent Muse Spark model, launched in April 2026 and internally codenamed Avocado, as merely an “appetizer” for what’s coming next.

From open-weight darling to proprietary contender

Meta’s Llama 4 release in April 2025 landed with a thud, drawing criticism that pushed leadership to rethink the entire approach. The shift from open-weight releases to a proprietary model strategy marked a philosophical U-turn for a company that had previously positioned itself as the champion of open AI development.

Wang joined Meta in June 2025, following a $14.3 billion investment in his previous company, Scale AI. Under his direction, the focus has shifted toward building scalable AI systems designed to be woven directly into Meta’s apps and hardware ecosystem.

Watermelon: Meta’s answer to GPT-5.5

The next model in Meta’s pipeline carries the codename Watermelon. According to reports, it reportedly approaches performance parity with OpenAI’s GPT-5.5, though it achieves this by throwing significantly more computational resources at the problem.

Meanwhile, Meta released Muse Spark version 1.3 in September 2026, which brought improvements to coding performance and overall efficiency.

Wang has characterized the upcoming releases as “highly competitive” against the best models currently available.