ResearchBenchmarksUser Simulation Quality Assessment on Turn-level Human Evaluation Set adversarial (test)Follow168Win RateUserLM-R162.9690.23117.5144.77Jan 14, 2026Evaluation ResultsMethodMethodLinksWin RateTie RateLoss RateUserLM-R1Opponent=DeepSeek-R1Opponent=DeepSeek-R12026.011682042UserLM-R1Opponent=Gemini-2.5-FlashOpponent=Gemini-2.5-Flash2026.011424038UserLM-R1Opponent=UserLM-R1 (w/...Opponent=UserLM-R1 (w/o RL)2026.016711142