Go Create Your Own Benchmark

Nico's advice to anyone doing vibe coding: go create your own benchmark. Pick a configuration, put together an app build package, hand the same package to every model at whatever level, and look at what comes back. It does not have to be complex. A working clock. A sextant you can move with your mouse that obeys physics. In his words, those are the ways you can consistently test how well these models do.