Daily AI CatchupMeta's In-Training 'Watermelon' Model Reportedly Matches GPT-5.5
Meta's superintelligence chief Alexandr Wang says the company's next frontier model, codenamed Watermelon, has caught up with OpenAI's GPT-5.5 on major benchmarks while still in training — using roughly 10x the compute of predecessor Muse Spark. No release timeline has been shared, but the benchmark parity claim signals Meta's frontier ambitions are producing real results.
Why it matters
💻 Developer · No API access yet, but if Watermelon ships at GPT-5.5 parity, it's worth planning for a third viable frontier option beyond OpenAI and Anthropic.
📦 Product · Meta closing the capability gap changes the competitive landscape you should be scenario-planning against for any AI-dependent product roadmap.
🎨 Design · Not yet actionable, but a third frontier-capable lab means more model choice for design-tool integrations down the line.
📈 Business · Meta's $145B+ AI infrastructure bet finally producing frontier-competitive results is the clearest signal yet that the spend is translating into capability, not just compute.
🤔 Just Curious · Meta says its next big AI model — still being trained — already performs as well as OpenAI's latest, suggesting Meta is catching up in the AI race.