Meta's upcoming 'Watermelon' AI model reportedly matches OpenAI's GPT-5.5 on key benchmarks

What happened

Meta AI chief Alexandr Wang announced at an internal town hall that the new 'Watermelon' model has reached GPT-5.5-comparable performance on key benchmarks, using an order of magnitude more compute. This is the first internal acknowledgment that Meta is closing the performance gap with OpenAI.

Context and impact

Benchmarks are internal and have not been independently verified. Watermelon follows previous Meta models Avocado and Muse Spark; a Muse Spark coding/agentic update is expected in the coming weeks. Wang's announcement comes as META stock faces pressure following Zuckerberg's separate admission of stalled agentic AI progress.

Details

• 'Watermelon' internally presented as GPT-5.5 peer on benchmarks (not independently verified)
• Alexandr Wang (Chief AI Officer) led the town hall presentation
• Training compute was 'an order of magnitude greater' than prior Meta models
• Muse Spark coding/agentic update is separate and imminent; Watermelon is an internal research project
• Model not yet publicly available; no public release date announced