Meta Says Its 'Watermelon' Model Has Caught Up With GPT-5.5
Meta's superintelligence chief Alexandr Wang told employees at a recent internal town hall, per Business Insider reporting published July 2 and picked up by Benzinga, that Watermelon, the company's next frontier model, has caught up with OpenAI's GPT-5.5 on certain benchmarks. Watermelon is still in training and reportedly uses an order of magnitude more compute than Avocado, per Business Insider; Avocado is Meta's internal codename for Muse Spark, the model family it released in April.
The caveats are doing heavy lifting. Wang did not name the benchmarks, Meta has not published any evaluation configuration, and the company declined to comment. CEO Mark Zuckerberg reportedly tempered the mood himself, saying the trajectory of agentic development over the past four months "hasn't really accelerated in the way that we expected." OpenAI, meanwhile, has already previewed the GPT-5.6 family, so the goalpost Wang cited may have moved before Watermelon ships.
The compute figure is the durable fact in the story. A tenfold jump over Muse Spark confirms Meta is still buying its way up the frontier curve, and internal benchmark claims are how labs justify that spend to their own staff before they can justify it to the market.