Meta's Upcoming 'Watermelon' AI Model Matches OpenAI's GPT-5.5 on Key Benchmarks, Alexandr Wang Reportedly Tells Employees

Meta Platforms Inc. (NASDAQ: META ) forthcoming AI model, Watermelon, has reportedly reached the same performance level as OpenAI’s GPT-5.5 model, as per the company’s Superintelligence Chief, Alexandr Wang.

Wang made this announcement during an internal town hall meeting, as reported by Business Insider on Thursday.

He claimed that the Watermelon model’s performance matches OpenAI’s GPT-5.5, based on certain AI model benchmarks.

The exact benchmarks Wang referred to are yet to be disclosed.

Wang said that Watermelon, Meta’s next AI model after Avocado (internally called Muse Spark), is currently being trained and requires significantly more computing power than Avocado.

Meta Platforms did not immediately respond to ‘s request for comments.

Read Also: OpenAI to Offer Trump Administration a 5% Equity Stake and Other AI Firms Could Also Follow: Report Meta AI Push Faces Zuckerberg’s Reality Check Wang’s evaluation comes amid CEO Mark Zuckerberg ‘s hefty investment and aggressive talent acquisition strategy in the heavily competitive AI landscape.

In April, Meta unveiled Muse Spark, its first major AI model since hiring Scale AI ‘s Alexandr Wang nine months ago.

Built by Meta Superintelligence Labs on a newly rebuilt AI stack, the model marks the company’s fastest AI development cycle to date.

While Meta is not positioning Muse Spark as a frontier model, it says the model delivers efficient, competitive performance and plans to open-source future versions of the Muse series.

Mizuho Securities views Meta’s early Muse Spark launch as a positive indicator of its AI competitiveness, noting that increased usage, particularly through Shopping mode, could significantly boost monetization via search and ad targeting.

However, during Thursday’s internal town hall, Zuckerberg struck a cautious tone and reportedly admitted that Meta’s AI bets have not materialized as quickly as anticipated.

He stated that the “trajectory of the agentic development over at least the last four months hasn’t really accelerated in the way that we expected.” He said leadership had been overly optimistic earlier this year, expecting AI tools, including coding assistants like Claude Code, to advance more quickly than they have.

Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published editors.

Image via Shutterstock Read Also: Meta Defends WhatsApp Usernames After India Raises Cybercrime Concerns and Seeks an Explanation