Meta claims new Watermelon model nearly matches OpenAI's GPT-5.5

Meta (Meta Platforms was recognized as an extremist organization in Russia, its activities in the country are banned) announced significant progress in the race for AI models: according to Alexander Wang (Alexandr Wang), head of Meta's AI division, the new system codenamed Watermelon has nearly caught up with OpenAI's flagship model GPT-5.5. This is reported by Business Insider, citing two sources.

What is known about Watermelon

The codename for the new model is Watermelon. The claim that it has nearly matched the capabilities of GPT-5.5 was made personally by Alexander Wang, who heads Meta's AI division. Business Insider cites two sources familiar with the situation, but does not name them. The company itself has not published an official statement on this matter — all information about Watermelon is based on Wang's words, retold by the publication citing anonymous sources.

Why the statement raises questions

The main problem with Wang's claim is that it is not backed by verifiable data. Business Insider notes: it was not possible to find out exactly what tests and benchmarks the comparison of Watermelon with GPT-5.5 is based on. This means that it is currently impossible to evaluate the claim independently — one can only rely on the words of Meta's AI division head. For an outside observer, this situation is typical of the early stages of the model race, when companies are more willing to share ambitions than reproducible test results.

• The codename for Meta's new model is Watermelon
• The statement was made by Alexander Wang, head of the company's AI division
• Business Insider published the information citing two sources
• Specific tests on which the comparison with GPT-5.5 is based have not been disclosed

Why codenames for models become news in the first place

Large AI labs regularly test new versions of models under internal codenames long before a public release — this is standard practice for OpenAI, Google DeepMind, and Meta alike. But the fact that a name like Watermelon leaks to the press along with a statement from a company leader about parity with the market leader turns the working codename into a media event: the market and investors interpret such leaks as a signal of real progress, even when benchmarks are not published.

Alexander Wang heads Meta's AI division and by position bears direct responsibility for ensuring the company does not fall behind OpenAI, Google, and other leaders in large language models. A public statement by a manager of this level about nearly achieved parity with GPT-5.5 is inevitably read in two ways: as an honest assessment of internal development progress or as part of a communication strategy designed to maintain Meta's reputation as one of the leaders in the race amid regular releases from competitors.

What this means

The competition between Meta and OpenAI is intensifying at the level of public statements: even without verified benchmarks, the very fact that Meta's AI division head openly speaks of parity with GPT-5.5 shows how important this competition is for the company's positioning before investors and the market. Until Watermelon passes independent verification and is released to the public, Alexander Wang's statement should be seen as a signal of Meta's internal ambitions rather than confirmation of verified technological parity with GPT-5.5.