Meta is claiming a major breakthrough in its AI development. Alexandr Wang, the company's chief of Superintelligence, announced internally that the new model, codenamed Watermelon, has matched OpenAI's GPT-5.5 on key benchmarks.
The announcement, made during a July 2 town hall, came with significant caveats. The benchmarks are internal and unnamed, and no independent verification exists. The model is also still in training.
Watermelon reportedly uses ten times the computational resources of its predecessor, the Avocado model. Meta's massive user base across its platforms provides a potential distribution advantage over competitors.
Investors are advised to exercise caution. The absence of public, replicable benchmarks is a critical detail. Industry observers note that companies often select metrics that present their models in the most favorable light.