Get Zlio on your home screen
One tap, full screen, like an app.
A whole-product benchmark shows coding agents still fail most end-to-end builds.
Replit released ViBench 2.0, a benchmark testing AI agents building 17 products from scratch and checking finished apps in a browser; top models pass 49%-65% of whole products.