Meta Says Its Watermelon AI Model Matches OpenAI GPT-5.5 on Key Benchmarks

Meta’s superintelligence chief, Alexandr Wang, told employees that the company’s next AI model has reached a new milestone in its effort to compete with leading AI systems. During an internal town hall, Wang said Meta’s upcoming model, codenamed Watermelon, has caught up with OpenAI’s flagship GPT-5.5 model based on widely watched AI benchmarks, according to two people familiar with the discussion. He did not specify which benchmarks he was referring to.

According to one person familiar with the meeting, Wang told employees that Watermelon is now in training and succeeds the internal project known as Avocado. He also said the new model is being trained with roughly an order of magnitude more compute than Avocado, Meta’s internal codename for Muse Spark, the first model in a family the company introduced in April.

Wang has also hinted publicly that Meta is preparing additional AI model updates. In a post on X, he said a new version of Muse Spark is coming soon with major improvements in coding and agentic capabilities intended to narrow the gap with competing models. When asked by a user when Meta would offer a coding model comparable to Anthropic’s Claude Opus, Wang responded that it would be “pretty soon,” adding that users would like what the company has “cooking.”

Meta has spent years trying to close the performance gap with OpenAI, Google, and Anthropic. Despite investing heavily in AI infrastructure and recruiting researchers, the company has faced questions about whether its models could consistently compete with the industry’s top systems.

If Wang’s internal assessment proves accurate, it would represent one of Meta’s strongest indications yet that its AI investments and recruiting efforts are translating into measurable model improvements, even as competitors continue releasing newer systems.

OpenAI released GPT-5.5 in April and later introduced GPT-5.6, although the newer model has not yet been made generally available following requests from the U.S. government. Meta’s Muse Spark models, released in April, delivered strong benchmark results but did not surpass leading models from OpenAI or Anthropic.

Meta CEO Mark Zuckerberg has made AI a central priority for the company. Last year, he appointed Wang to lead the effort and renamed the division Meta Superintelligence Labs.

Wang now oversees Meta’s elite AI research team known as TBD, along with other AI initiatives, including the company’s expanding hardware efforts. Business Insider previously reported that Meta has offered top AI researchers compensation packages worth hundreds of millions of dollars to join the company.

The hiring campaign has coincided with increased infrastructure spending. Meta told investors that it now expects to spend between $125 billion and $145 billion this year on chips, data centers, and other infrastructure, raising its previous forecast of $115 billion to $135 billion because of higher component costs and additional data center investments.

This analysis is based on reporting from Business Insider.

Image courtesy of Meta.

This article was generated with AI assistance and reviewed for accuracy and quality.