Meta’s “Watermelon”: New AI Model Reportedly Matches OpenAI’s GPT-5.5

The AI landscape is approaching another turning point. According to reports, Meta has made a significant technical leap with its latest model, “Watermelon.” While internal benchmarks suggest parity with OpenAI’s GPT-5.5, experts remain cautious given that the data has not yet been verified.

The race for dominance in artificial intelligence has reached a new level of intensity. In an internal town hall meeting, Alexandr Wang, Meta’s Superintelligence chief, reportedly told employees that the upcoming AI model, codenamed “Watermelon,” has already matched the performance of OpenAI’s flagship model GPT-5.5 on key benchmarks.

“Watermelon”: A Massive Bet on Compute

“Watermelon” is the direct successor to Meta’s “Muse Spark” model series (internally referred to as “Avocado”), which was released in April. The technological difference, however, is fundamental: according to Wang, Meta is training Watermelon with an order of magnitude (a factor of 10) more compute than its predecessor models.

This massive deployment of compute resources underscores Meta’s intention to no longer merely observe the technological frontier, but to actively shape it. The model is currently still in training, which makes it difficult to assess its performance timeline.

Meta’s New Strategy: From Follower to Frontier Setter

For a long time, Meta was seen as trailing OpenAI and Anthropic in AI development. The new strategy, however, appears clearly defined: Meta is betting on sheer scale. By investing in gigantic compute capacity, the company aims to close the gap to the “frontier models.”

The goal is to integrate this advanced AI deeply into Meta’s existing ecosystem – including Facebook, Instagram, and WhatsApp. Meta is thus pursuing a hybrid approach: developing cutting-edge models on the one hand, while making them immediately available to billions of users across its social networks on the other.

The Comparison: GPT-5.5 and Expert Skepticism

Whether “Watermelon” can live up to this claim is currently the subject of intense debate. A look at the Artificial Analysis ranking shows the current top of the industry: while GPT-5.5 (xhigh) ranks among the absolute frontrunners with 55 points on the Intelligence Index, Anthropic’s Claude Fable 5 sits even higher. Meta’s new LLM would therefore not be able to claim the top spot, but would likely land in the top 5. Not insignificant: Meta could beat Google’s current top models.

Nevertheless, the data reveals a nuanced reality that calls for caution:

1. Unverified data: The benchmarks Meta has cited for “Watermelon” are not yet publicly available and have not been validated by independent third parties.
2. Nuanced dominance: While GPT-5.5 remains extremely strong in technical metrics, the dynamic top of the leaderboards shows how tight the field is between OpenAI and Anthropic.
3. Version dynamics: While Meta is trying to catch up with GPT-5.5, OpenAI has already taken the next step. GPT-5.6 is already available to selected partners, meaning Meta is technically chasing a moving target.