ResearchBenchmarksCraft TaskFollow14.14ScoreCOMPACTOR8.711210.120611.5312.9394May 25, 2026Evaluation ResultsMethodMethodLinksScoreCOMPACTORModel=Qwen3 30B ThinkModel=Qwen3 30B Think2026.0514.14HumanModel=HumanModel=Human2026.0512.57HUMPRModel=Llama 3 8BModel=Llama 3 8B2026.058.92