UI-TARS-2 Technical Report: Advancing GUI Agent with Multi-Turn Reinforcement Learning

Benchmark Model Rank Results
computer-use-agents-on-online-mind2webUI-TARS-2#1Success Rate: 88.2
computer-use-agents-on-windowsagentarena-50-stepsUI-TARS-2#5Success Rate: 50.6
computer-use-agents-on-androidworldUI-TARS-2#12Success Rate: 73.3
code-generation-on-swe-bench-verifiedUI-TARS-2 (w/ GUI-SDK)#48Accuracy: 68.7
deep-research-agents-on-browsecompUI-TARS-2 (w/ GUI-SDK)#55Accuracy: 29.6