Meta says its new Watermelon model has nearly caught up with OpenAI's GPT-5.5

Meta (Meta Platforms is designated an extremist organization in Russia, and its activities are banned in the country) says it has made significant progress in the AI model race. According to Alexandr Wang, who heads the company's AI division, a new system codenamed Watermelon has nearly caught up with OpenAI's flagship model, GPT-5.5. Business Insider reported the news, citing two sources.

What we know about Watermelon

The new model's codename is Watermelon. Alexandr Wang, who leads Meta's AI division, personally made the claim that it has nearly matched GPT-5.5 in capabilities. Business Insider cites two sources familiar with the matter but does not name them. Meta itself has not published an official statement on the subject, so everything known about Watermelon so far rests on Wang's words as relayed by the publication through anonymous sources.

Why the claim raises questions

The main problem with Wang's claim is that no verifiable data backs it up. Business Insider notes that it could not determine which tests and benchmarks the Watermelon-versus-GPT-5.5 comparison is based on. That means the claim cannot yet be independently assessed; for now, the only thing to go on is the word of Meta's AI chief. To an outside observer, this is typical of the early stages of a model race, when companies are more willing to share ambitions than reproducible test results.

• Meta's new model is codenamed Watermelon
• The claim came from Alexandr Wang, head of the company's AI division
• Business Insider published the information, citing two sources
• The specific tests behind the comparison with GPT-5.5 have not been disclosed

Why model codenames make news at all

Major labs routinely test new model versions under internal codenames long before public release. It's standard practice at OpenAI, Google DeepMind, and Meta alike. What turns a working codename into a media event is when a name like Watermelon leaks to the press alongside a top executive's claim of parity with the market leader. The market and investors read such leaks as a signal of real progress, even when no benchmarks have been published.

Alexandr Wang runs Meta's AI division, and his job makes him directly responsible for keeping the company from falling behind OpenAI, Google, and the other leaders in large language models. A public claim from an executive at his level that the company has nearly reached parity with GPT-5.5 inevitably reads two ways: as an honest assessment of internal development progress, or as part of a communications strategy meant to shore up Meta's reputation as a frontrunner while competitors keep shipping releases.

What it means

The Meta–OpenAI race is heating up at the level of public statements. Even without confirmed benchmarks, the fact that Meta's AI chief is openly talking about parity with GPT-5.5 shows how much this rivalry matters to the company's positioning with investors and the market. Until Watermelon passes independent evaluation and gets a public release, Alexandr Wang's claim should be read as a signal of Meta's internal ambitions, not as confirmed proof of technological parity with GPT-5.5.