I gave 9 free AI coding models the same real bug. Here is what happened.

Same open source bug, same brief, hidden tests the models never saw. 8 of 9 free models fixed it, the fastest in 84 seconds. The interesting part was how the others went wrong.