ResearchBenchmarksPearson Correlation with Human Evaluation on HOWTOBENCH EnglishFollow0.85Completion CorrelationROUGE-L0.6420.6960.750.804Apr 21, 2026Evaluation ResultsMethodMethodLinksCompletion CorrelationGuide CorrelationOpenness CorrelationOverall CorrelationROUGE-L2026.040.850.340.480.52ToW2026.040.820.820.860.85Rubrics2026.040.760.850.810.81BLEU-12026.040.740.650.690.71Auto Plan2026.040.650.710.540.69