ResearchBenchmarksConversation Simulation on GSM8k and HumanEval 65 task intentsFollow86.6Intent Coverage4o-mini76.30478.97781.6584.323Oct 8, 2025Evaluation ResultsMethodMethodLinksIntent CoverageRepeat Rate (%)Non-Required Skip Rate (%)Added Demand Rate (%)Turn VarianceMinimum TurnsUnigram DifferenceAssistant Score4o-miniUser Simulator=4o-miniUser Simulator=4o-mini2025.1086.631.810.99.50.93.70.4373.2GPT-4oUser Simulator=GPT-4oUser Simulator=GPT-4o2025.1084.79.414.61.10.640.474.6UserLM-8bUser Simulator=UserLM-8bUser Simulator=UserLM-8b2025.1076.754.337.743.82.82.10.7157.4