ResearchBenchmarksReasoning Trace Quality Evaluation on DROPFollow2.8GrammarCRAFT1.3441.7222.12.478Apr 15, 2026Evaluation ResultsMethodMethodLinksGrammarRep-StepRep-WordCRAFTModel=o4-miniModel=o4-mini2026.042.82.72.2CRAFTModel=GPT-5.4-nanoModel=GPT-5.4-nano2026.041.41.82.1