Meta's 'Watermelon' AI Model Has Caught Up to GPT-5.5, Exec Says

Meta’s next-generation AI model, codenamed Watermelon, has matched OpenAI’s GPT-5.5 in key benchmarks — a sign that the social media giant’s massive AI investment is starting to pay off.

Wang Tao, who leads Meta’s super-intelligence division, shared the news with employees during an internal all-hands meeting, according to two people familiar with the discussion. Wang did not specify which benchmarks the model was tested on.

“Watermelon is the next-generation model after Avocado, and it’s still in training,” Wang said at the meeting, according to one of the people. “Watermelon uses an order of magnitude more compute than Avocado did.”

Wang has also been public about the progress. In a post on X, he said Meta’s Muse Spark model would soon receive updates that significantly improve its coding and agent capabilities, narrowing the gap with competitors. When a user asked when Meta might ship a coding model that matches Anthropic’s Claude Opus, Wang replied: “Soon. You’ll like what we have next.”

If Wang’s assessment holds, it signals that Meta’s aggressive spending and talent war — led by CEO Mark Zuckerberg — are producing results. OpenAI released GPT-5.5 in April, followed by the even more capable GPT-5.6 at the end of last month, though the US government has asked OpenAI to hold off on a full public release of that model.

Meta launched its first Muse Spark model in April. It performed well in benchmarks but didn’t close the gap with labs like OpenAI and Anthropic.

Zuckerberg has made AI leadership a top priority. Last year, he put Wang in charge of the effort and rebranded Meta’s AI division as the Super Intelligence Lab. Wang now leads a top-tier AI research team called TBD and oversees hardware and other AI projects. Reports have circulated that Meta has offered top AI researchers compensation packages worth hundreds of millions of dollars each to join the company.

The talent war is expensive, and so is the infrastructure. Meta told investors this year it expects to spend between $125 billion and $145 billion on chips, data centers, and other infrastructure — up from an earlier forecast of $115 billion to $135 billion, driven by rising component costs and increased data center spending.