@millions-crayon-32880 I think that the recalculate-cost workflow was not counting critic runs, which caused it to undercount cost a little. and also one run of qwen-3.5-flash had wrong costs? here’s a fix • https://github.com/OpenHands/openhands-index-results/pull/908 • https://github.com/OpenHands/benchmarks/pull/694 • https://github.com/OpenHands/evaluation/pull/525