Meta's Watermelon Matches GPT-5.5, Wang Tells Staff
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Alexandr Wang, Meta's superintelligence chief, told employees that Watermelon — the model currently in training — matches OpenAI's GPT-5.5 and uses an "order of magnitude more compute than Avocado," Meta's previous AI model
- Mark Zuckerberg told staff at a town hall that AI agent development has not accelerated as expected and that the company's reorganization was not as "clean" as it could have been
- Watermelon's reported 10x compute increase over Avocado comes as Implicator.ai frames Meta's 8,000-person layoff-driven reorg as having failed to deliver, even as the next model hits GPT-5.5 parity
- Wang also teased an upcoming Muse Spark update with improvements in coding and agentic capabilities, alongside an Opus-level Muse variant, per Andrew Curran's reporting of the same internal meeting
- Charles Rollet, who broke the Watermelon scoop, noted Wang's framing distinguished the model's benchmark catch-up from the broader industry's agentic progress — a hedge Zuckerberg's own remarks undercut
Why it matters: Meta's top AI leaders are sending the market contradictory signals within the same week: Wang claims the next flagship model reaches GPT-5.5 parity, while Zuckerberg admits the agentic work and 8,000-person reorg are underperforming. The order-of-magnitude compute escalation from Avocado to Watermelon implies Meta is willing to absorb sharply higher infrastructure costs to close the frontier gap with OpenAI, even as internal confidence on execution wavers.
