During an internal gathering, Alexandr Wang, Meta's head of superintelligence, revealed substantial advancements in the company's efforts within the AI sector. He informed attendees that Meta's forthcoming AI model, referred to as Watermelon, is now on par with OpenAI's prominent GPT-5.5 model, according to sources familiar with the situation. Wang highlighted this progress based on AI model benchmarks, though specific details regarding these benchmarks were not disclosed.
"Watermelon, our successor to Avocado, is actively in training," Wang shared during the town hall, as reported by an insider. He also noted that "Watermelon leverages significantly more computational power compared to Avocado," mentioning that Avocado is the internal codename for Muse Spark, the inaugural model from a series released by Meta in April.
The core objective behind Meta's AI initiatives has consistently been to bridge the gap with leading organizations like OpenAI, Google, and Anthropic. Although the company has heavily invested in advanced chips, data centers, and expert talent, it has faced challenges in demonstrating that its models are competitive at the industry's forefront.
If Wang's statements hold true, this could signify a pivotal moment indicating that Meta's substantial investments and CEO Mark Zuckerberg's aggressive talent acquisition strategies are beginning to yield results, even as the competitive landscape continues to evolve rapidly.
OpenAI's GPT-5.5, a powerful chatbot, was introduced in April, and the organization subsequently unveiled its most advanced model to date, GPT-5.6, last month. However, general access to GPT-5.6 has not been made available yet, pending requests from the U.S. government.
When approached for comments, Meta declined, and OpenAI did not respond to inquiries.
In April, Meta launched its initial model in the Muse Spark series, achieving favorable results on benchmarks, yet not surpassing the performance of OpenAI or Anthropic's offerings.
Zuckerberg is determined to propel Meta to the forefront of the AI landscape. Last year, he appointed Wang to spearhead this initiative, rebranding the company's AI division as Meta Superintelligence Labs.
Wang supervises a distinguished team of AI researchers known as TBD, alongside other AI-related projects, including a recent initiative aimed at enhancing hardware capabilities. Reports from Business Insider indicate that Meta has been providing lucrative offers to top AI talent, some totaling hundreds of millions of dollars each.
This talent acquisition strategy coincides with Meta's increased investments in infrastructure. The company recently informed investors that it expects its spending on chips, data centers, and related infrastructure to reach between $125 billion and $145 billion this year, a revision from its earlier estimate of $115 billion to $135 billion due to rising component prices and additional data center costs.
