Meta's head of superintelligence informed staff on Thursday that the company has made significant strides in the artificial intelligence model competition.
According to individuals familiar with the matter, Alexander Wang stated in an internal meeting that Meta's upcoming AI model—codenamed "Watermelon"—has matched the performance of OpenAI's flagship model, GPT-5.5. Wang cited industry-closely-watched AI model benchmark data to substantiate this claim. It remains unclear which specific evaluation standards were referenced.
One source revealed that Wang told the meeting, "Following 'Avocado,' our next model 'Watermelon' is currently in training." He added, "'Watermelon' has a computational capacity an order of magnitude greater than 'Avocado.'" Wang was referring to "Avocado," which is Meta's internal codename for Muse Spark, the first in a series of models the company released in April.
Meta's artificial intelligence ambitions have long hinged on a straightforward goal: closing the gap with OpenAI, Alphabet (parent of Google), and Anthropic. Despite substantial investments in chips, data centers, and talent, the company has struggled to convince developers and clients that its models are industry-leading.
If Wang's assessment proves accurate, it would mark the clearest sign to date that Meta's investments and Mark Zuckerberg's aggressive talent blitz are beginning to yield returns, even as the competitive landscape continues to evolve rapidly.
GPT-5.5 is a powerful chatbot released by OpenAI in April of this year. At the end of last month, OpenAI introduced its most powerful model to date, GPT-5.6, though it has not been officially released yet due to requirements from the U.S. government.
Comments