GUI Agent Task Success on OSWorld
24.4Success RateUI-TARS-7B-DPO + ROSA2
Evaluation Results
| Method | Links | |
|---|---|---|
| UI-TARS-7B-DPO + ROSA2Backbone=UI-TARS-7B-DPO2026.03 | 24.4 | |
| UI-TARS-7B-SFT + ROSA2Backbone=UI-TARS-7B-SFT2026.03 | 23.6 | |
| UI-TARS-7B-DPO + ROSABackbone=UI-TARS-7B-DPO2026.03 | 18 | |
| UI-TARS-7B-SFT + ROSABackbone=UI-TARS-7B-SFT2026.03 | 17.8 | |
| UI-TARS-7B-DPO + TextGradBackbone=UI-TARS-7B-DPO2026.03 | 14.9 | |
| UI-TARS-7B-DPOBackbone=UI-TARS-7B-DPO2026.03 | 14.8 | |
| UI-TARS-7B-SFT + TextGradBackbone=UI-TARS-7B-SFT2026.03 | 13.7 | |
| UI-TARS-7B-SFTBackbone=UI-TARS-7B-SFT2026.03 | 13.2 | |
| ShowUI-AlohaType=Human Guided2026.01 | 0.601 | |
| CoAct-1Type=Agentic2026.01 | 0.564 | |
| Agent S2.5Type=Agentic2026.01 | 0.542 | |
| Jedi-7BType=Agentic2026.01 | 0.506 | |
| GTA-1-7BType=Agentic2026.01 | 0.486 | |
| Claude 4 SonnetType=General2026.01 | 0.439 | |
| OpenAI CUA 4oType=Specialized2026.01 | 0.313 | |
| UI-TARS-1.5-7BType=Specialized2026.01 | 0.294 |