Meta's Wang Says Watermelon Model Has Caught Up to GPT-5.5

Meta Superintelligence Labs chief Alexandr Wang reportedly told an internal town hall that the company's still-in-training next flagship, codenamed Watermelon, has caught up with OpenAI's GPT-5.5 on closely followed benchmarks, according to Business Insider. He did not say which benchmarks. Meta declined to comment and OpenAI did not respond.

The shape of the claim matters. Wang described Watermelon as the successor to Avocado, the internal codename for Muse Spark, the model family Meta released in April. Watermelon, per Wang, uses an order of magnitude more compute than Avocado. So the story is not that Meta found a clever architectural trick that closed the gap, it is that Meta poured a lot more compute into a bigger training run and reached parity on benchmarks with a model, GPT-5.5, that OpenAI itself shipped in April and has already partly moved past with a late-June limited preview of GPT-5.6.

That framing is what makes this interesting for practitioners rather than just Meta-watchers. Muse Spark performed well on benchmarks but did not match or exceed OpenAI or Anthropic, while Meta now projects $125 billion to $145 billion in 2026 spending on chips, data centers, and infrastructure, up from an earlier $115 billion to $135 billion range. If Watermelon ships close to what Wang described, that spending starts to look like something other than expensive catching up. Wang has also hinted on social media that a Muse Spark update with better coding and agentic capabilities is coming, and asked about competing with Anthropic's Claude Opus for coding, he replied it would be 'pretty soon.'

The honest caveat is that all of this is one executive, in one internal room, on one round of unnamed benchmarks, with no confirming third-party evaluation and no comment from OpenAI. Take the specifics as reported, not settled. What the reporting does not give you is a ship date for Watermelon, the actual benchmark table, or any evaluation on the reasoning and agentic tasks where Anthropic's Opus line currently sets the bar.

If Wang's claim survives contact with a public release and independent evals, the near-term beneficiaries are enterprise buyers currently locked into a two-vendor OpenAI-and-Anthropic frontier, and Meta's own ads and product teams that need a competitive in-house model. If it does not, the town hall quote will read as the most expensive morale slide of the year.