David Kennedy

As someone who has spent their career breaking into systems and defending against attacks, I've learned one thing: you can't defend what you can't measure. The same principle applies to AI models.

I started noticing that frontier AI models — the ones we rely on for coding, analysis, and security work — would silently degrade without any announcement. One day a model is exceptional at finding bugs in code, and the next week it's introducing them. Providers don't always communicate these changes, and by the time you notice, you've already shipped code reviewed by a model that was performing below its previous standard.

This project exists because the AI community deserves transparency. ModelRegression.com runs automated benchmarks every day against every major frontier model, tracking performance across real-world tasks: coding, bug detection, security analysis, reasoning, and more.

When a model regresses, you'll see it here — with evidence. When it recovers, you'll see that too. The goal is simple: give developers and teams the data they need to make informed decisions about which model to use today, not which model was best last month.