OpenHands

This seems something that worth running OpenHands against :thinking_face:
<https://programbench.com/>

<https://github.com/OpenHands/benchmarks/tree/main/benchmarks/programbench|benchmarks/benchmarks/programbench at main · OpenHands/benchmarks>

Not super-sure that i think the benchmark is super high-quality though, you kinda need to guess the requirements which are pretty underspecified. It's not a very realistic situation

FWIW we got "close" with Claude Sonnet 4.5 on one instance, and failed 2