ResearchBenchmarksSimulated Manipulation on LIBEROFollow97.8Success RateKeyStone48.50461.30274.186.898May 9, 2026Evaluation ResultsMethodMethodLinksSuccess RateDeltaKeyStoneModel=π0.5, K=16Model=π0.5, K=162026.0597.81π0.5K=1K=12026.0596.8—KeyStoneModel=SmolVLA, K=16Model=SmolVLA, K=162026.0557.26.8SmolVLAK=1K=12026.0550.4—