I benchmarked Jev against three cheap chat models, then had to publish two corrections