- OSworld (100 steps) 42.5
- Windows Agent Arena (50 steps) 42.1
- WebVoyager 84.8
- Online-Mind2web 75.8
- Android World 64.2
- ScreensSpot-V2 94.2
- ScreenSpotPro 61.6
UI-TARS-1.5
UI-TARS-1.5 places in the top three on 2 benchmarks.
15
Benchmarks with a score
0
Ranked first, out of five or more models
2
In the top three
ByteDance Seed Apr 2025 UI-TARS-1.5
Closed weights Mid-size
No Epoch Capabilities Index score yet
- GameWorld 31.1% #21 /22
- ScienceBoard 2.69% #14 /15
- WARC-Bench 10.3% #6 /8
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ScreenSpot-v2 | 94.2% | #3 of 14 | Lab model card | |
| OSWorld | 42.5% | #5 of 13 | Lab model card | |
| Windows Agent Arena | 42.1% | #2 of 5 | Lab model card | |
| WebVoyager | 84.8% | #6 of 12 | Lab model card | |
| AndroidDaily | 47% | #1 of 2 | Official leaderboard | |
| Online-Mind2Web | 75.8% | #6 of 10 | Lab model card | |
| AndroidWorld | 64.2% | #14 of 23 | Lab model card | |
| ScreenSpot-Pro | 61.6% | #22 of 32 | Lab model card | |
| WARC-Bench | 10.3% | #6 of 8 | Official leaderboard | |
| OSWorld-Verified | 27.4% | #30 of 33 | Official leaderboard | |
| ScienceBoard | 2.69% | #14 of 15 | Official leaderboard | |
| GameWorld | 31.1% | #21 of 22 | Official leaderboard | |
| GameHorizon | 53.0 | #37 of 37 | Official leaderboard | |
| MobileGym | 13.8% | #4 of 4 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 43.9 | #174 of 185 | Official leaderboard |
Its lab's number next to a leaderboard
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| ScreenSpot-Pro | 61.6% ScreenSpotPro | 61.6% Leaderboard | Same |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.