running some updated model benchmark evals (now using the latest benchmarks repo as a base instead of an old fork) 🙂 would love a review on https://github.com/OpenHands/benchmarks/pull/522