Meta's Watermelon reportedly matches GPT-5.5 internally, uses 10x compute
Meta's new AI model, Watermelon, is making waves. It's reportedly performing as well as OpenAI's GPT-5.5 in internal tests.
Meta Superintelligence Labs chief Alexandr Wang shared this update at a recent town hall.
Watermelon isn't finished yet and uses 10 times more computing power than the previous model, Avocado (also known as Muse Spark).
Mark Zuckerberg cites $145B infrastructure plan
Meta is going all in to catch up with rivals like OpenAI and Anthropic, pouring resources into AI research and talent.
Mark Zuckerberg admitted progress has been slower than hoped, even with plans to spend up to $145 billion on AI infrastructure this year.
He also talked about ongoing company restructuring, including recent layoffs, to help Meta move faster in the changing world of AI.
Expect major upgrades to Muse Spark soon and broader improvements across Meta's AI systems within the next few months.