Last updated: 7 July 2026

Meta's AI Chief Says an Unreleased Model Matches GPT-5.5

At an internal town hall on 2 July 2026, Meta Superintelligence Labs chief Alexandr Wang told employees that the company's next model, codenamed Watermelon, has caught up with OpenAI's GPT-5.5 on closely followed benchmarks, according to Business Insider, citing two people familiar with the meeting. Neither Meta nor OpenAI has confirmed the claim, and Business Insider reported it is not clear which benchmarks Wang cited.

Figures labeled claim are operator-stated or reported and not independently audited.

What Wang reportedly said

Wang told the town hall that Watermelon "uses an order of magnitude more compute than" Muse Spark and that it has closed the gap with GPT-5.5, OpenAI's flagship model released in April, according to Business Insider's sourcing. That reporting has since been repeated by Reuters and more than a dozen other outlets, but all trace back to the same original story, built on anonymous accounts of a private meeting. No published benchmark table, methodology, or independent evaluation supports the claim yet.

Zuckerberg's caution, and Wang's clarification

At the same meeting, Zuckerberg reportedly said Meta's AI agent work "hasn't accelerated in the way we expected" and that the company's recent restructuring, which included about 8,000 layoffs, "haven't come to fruition yet," per Reuters. He said he expects a payoff within three to six months.

Wang addressed the apparent tension directly on X the same day, writing that Zuckerberg was referring to the industry's agent progress broadly, not Meta's specifically. In the same post, Wang confirmed a Muse Spark update is coming soon with "big improvements in coding and agentic capabilities," rolling out to both Meta AI and a new developer API.

The coding question

Asked on X when Meta would ship a coding model comparable to Anthropic's Claude Opus, Wang responded that it would happen "pretty soon," according to Tekedia's account of the exchange; other outlets described his response more loosely as a commitment to closing the gap, without quoting that exact phrase. Wang has not stated a firm date. Claim

Why this is hard to verify right now

Every report of the benchmark claim traces to one Business Insider story built on anonymous sources, not ten independent ones. Wang's own X posts are the only on-record, primary material available, and those confirm the Muse Spark update and his characterization of Zuckerberg's remark, not the GPT-5.5 parity claim itself. Meta has previously released benchmark data for Muse Spark at launch in April; whether it does the same for Watermelon will be the real test of Wang's claim.

This article covers claims made in a private meeting and on social media that have not been independently verified or confirmed by Meta or OpenAI. We will update this piece if Meta publishes benchmark data for Watermelon.

Last updated: 7 July 2026