Benchmarks you can actually check.
An open benchmark for coding-agent harnesses, so anyone can check if a new version really got better.
95%
intervals on every score, not one lucky run
Product Design · Web · Design System · Brand
(01) Projects
Bring the problem. We'll bring the rest.
We take on a few new products each quarter. Tell us what you're building and who it's for.
2 days
to reply with a first take
Web · Mobile · Internal Tools