Alexandr Wang Says Meta Has Closed the Gap

Meta's chief AI officer, Alexandr Wang, told employees during an internal town hall that the company's next model has caught up with OpenAI's GPT-5.5 on some of the industry's most closely watched benchmarks. That's according to Business Insider, which first reported the comments. If it holds up, it's the strongest sign yet that Meta's massive AI spending is starting to pay off against the field's frontrunners.

What Wang Actually Told the Room

According to a person familiar with the matter, Wang described the model — internally code-named Watermelon — as the successor to Avocado, Meta's prior in-house project name. He said Watermelon is still in training and that it's burning through roughly ten times the compute of its predecessor, Muse Spark, which is the public name Meta gave to the model once known internally as Avocado when it launched back in April.

Wang didn't say which specific benchmarks he was referencing, and there's been no independent confirmation of the claim so far. Meta declined to comment on the record, and OpenAI didn't respond to a request for comment either.

Why GPT-5.5 Might Already Be an Outdated Target

Here's the catch: GPT-5.5, the model Wang used as his comparison point, came out back in April. OpenAI has already moved past it, quietly debuting GPT-5.6 late last month as its most capable model yet. That version hasn't rolled out broadly, reportedly at the request of the U.S. government. So even if Watermelon really does match GPT-5.5, Meta could still be chasing a target that's already moved further down the road.

Avocado's Bumpy Road to Becoming Muse Spark

This isn't the first time Meta's AI timeline has slipped. Avocado was originally slated to ship in late 2025, but it got pushed back more than once after internal testing turned up weak spots in reasoning, coding, and writing compared to what rivals were putting out. By the time it finally shipped in April under the public name Muse Spark, it still didn't outperform the models coming out of OpenAI or Anthropic.

Zuckerberg's Multibillion-Dollar Bet on Catching Up

None of this is happening on a shoestring budget. Mark Zuckerberg has committed serious money to closing Meta's AI gap, with the company expecting to spend somewhere between $125 billion and $145 billion this year alone on chips, data centers, and other infrastructure. Part of that bet was bringing in Wang himself, the founder of Scale AI, to lead Meta Superintelligence Labs starting in mid-2025 — reportedly with pay packages worth hundreds of millions of dollars on offer to pull in top researchers.

The Finish Line Keeps Moving

Watermelon is still in training, so there's no way to know yet whether it'll actually deliver once it's out in the world. That's really the nature of this race right now: by the time any one company closes the gap, the target usually has already shifted again.