Meta’s Upcoming ‘Watermelon’ AI Model Matches GPT-5.5 Benchmarks, Says Chief AI Officer Alexandr Wang

Meta is making another major push in the artificial intelligence race, with the company’s Chief AI Officer, Alexandr Wang, revealing that its upcoming AI model, codenamed Watermelon, has achieved performance comparable to OpenAI’s GPT-5.5 on key internal benchmarks.

According to a report by Business Insider, Wang shared the update during an internal town hall meeting, telling employees that the next-generation model is still undergoing training but has already reached a level that rivals one of the industry’s most advanced AI systems. While he did not disclose which benchmarks were used or provide detailed evaluation results, the announcement highlights Meta’s growing ambition to compete at the forefront of generative AI.

Watermelon is expected to succeed Meta’s previous flagship frontier model, Avocado, which was publicly released earlier this year under the name Muse Spark. The new model is reportedly being trained with approximately ten times more computing power than its predecessor, reflecting Meta’s massive investment in AI infrastructure and research.

Wang also revealed that future versions of Watermelon are expected to deliver significant improvements in coding capabilities and AI agent performance, two areas that have become increasingly competitive among leading AI developers such as OpenAI and Anthropic.

The update comes just days after Meta CEO Mark Zuckerberg acknowledged that the company’s progress in building advanced AI agents has been slower than initially anticipated. Speaking during a separate internal meeting, Zuckerberg admitted that Meta underestimated the time required to reorganize its AI operations and integrate new research efforts into practical products.

Despite the slower-than-expected progress, Zuckerberg expressed confidence that Meta’s substantial AI investments would begin delivering stronger results over the next three to six months as several new projects move closer to launch.

Meta’s AI development is now being led by its newly established Superintelligence Labs, a division created to accelerate the company’s frontier AI research and strengthen its position against competitors including OpenAI and Anthropic. Alexandr Wang took charge of the division after Meta acquired his AI startup, Scale AI, in a move widely seen as a strategic effort to attract top AI talent and accelerate innovation.

While Meta’s claims about Watermelon’s performance have yet to be independently verified, the announcement signals that competition among the world’s leading AI companies is intensifying. If the model lives up to expectations, Watermelon could become one of Meta’s most powerful AI systems to date, further fueling the race to build increasingly capable and intelligent AI models.