ResearchBenchmarksOpen-ended tasks on MT-Bench (test)For AI assistantsMarkdownContent indexOpen-ended tasks on MT-Bench (test)Follow44.8Win RateCOPE42.40843.02943.6544.271Jun 13, 2025Evaluation resultsMethodMethodLinksWin RateCostCOPESmall Model=EXAONE-3.5...Small Model=EXAONE-3.5-2.4B-Instruct, Large Model=GPT-4o-mini, Judge Model=GPT-4o2025.0644.88.64LargeModel=GPT-4o-miniModel=GPT-4o-mini2025.0642.511.06