Meta's AI Chief Says Its New Model 'Watermelon' Has Caught Up to GPT-5.5

Meta may be closing the gap in the AI race — at least by its own account. At an internal town hall, the company's superintelligence chief, Alexandr Wang, told staff that its still-in-training model, codenamed Watermelon, has caught up to OpenAI's flagship GPT-5.5 on major benchmarks. Per Business Insider, Wang didn't say exactly which tests or what results he meant.

In brief

• Whose claim: Meta superintelligence chief Alexandr Wang, at an internal town hall (early July 2026). No independent verification or disclosed scores.
• What Watermelon is: Meta's next model, the successor to "Muse Spark" (an April release, internally codenamed Avocado). Still in training.
• Scale: Watermelon is being trained with roughly 10x more compute than Muse Spark.
• Also coming: a Muse Spark update with stronger coding and agent capabilities, "pretty soon."

Our take

First, a caveat: there's no public model and no disclosed score here — Meta didn't say which test or what result, and the model is still training. So "caught up" is early. Still, the real meaning is the race getting crowded: it's no longer just Anthropic and Google chasing GPT-5.5 — Meta is in it too. That's good for you; as strong frontier models multiply, you get more choice and downward pressure on prices. For now the only thing to do is keep this race on your radar; there's no reason yet to run and switch models.

Kaynak: Business Insider