Meta's 'Watermelon' Reportedly Matches GPT-5.5 — By Burning an Order of Magnitude More Compute

Meta's next frontier model is said to reach GPT-5.5-level performance on internal benchmarks while consuming roughly 10x the compute of its predecessor.

Meta's forthcoming frontier model, code-named "Watermelon," matches GPT-5.5 on internal benchmarks, according to a roundup from @TraffAlex. The more revealing detail is the cost of getting there: Watermelon reportedly uses "an order of magnitude more compute" than Meta's previous "Avocado" model.

Unlock the full briefing

Get every story in today's briefing, the full archive, and the daily AI intelligence brief.

All stories today

Full archive

Daily brief

Cancel anytime. Payments powered by Stripe.