uitars-1.5-7B: benchmark results
ByteDance's UI-TARS-1.5-7B, an open multimodal GUI agent that reasons via reinforcement learning before acting, built for computer-use and virtual-environment tasks. Provider: ByteDance. Released 2025-04-16. Access: Open.
Aggregate Rating 1476 ± 12, rank #864 of 1656 rated models, from 97 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AgentHijack | 18.74 | Task Success Rate (%; mean of the clean run and the nine cor | 100 |
| AgentHijack - Accidental Touch | 22.54 | Task Success Rate (%; accidental click at step 3) | 100 |
| AgentHijack - App Minimization | 20.84 | Task Success Rate (%; application minimized at step 3) | 100 |
| AgentHijack - Clean | 24.21 | Task Success Rate (%; OSWorld tasks without corruption) | 100 |
| AgentHijack - Multiple Apps | 19.25 | Task Success Rate (%; a second application window open) | 100 |
| AgentHijack - Network Error | 22.02 | Task Success Rate (%; network disconnected) | 100 |
| AgentHijack - Screen Marks | 23.31 | Task Success Rate (%; 50 marks drawn over the screen) | 100 |
| AgentHijack - Subtitle | 22.75 | Task Success Rate (%; subtitle overlay) | 100 |
| AgentHijack - Verification | 10.48 | Task Success Rate (%; login verification demanded) | 100 |
| DocOS - Easy | 30.86 | Task completion rate (%; share of tasks whose final applicat | 100 |
| DocOS - Hard | 13.08 | Task completion rate (%; share of tasks whose final applicat | 100 |
| DocOS - Medium | 17.32 | Task completion rate (%; share of tasks whose final applicat | 100 |
Interactive version: theaggregate.ai/model?slug=uitars-1-5-7b · How It Works · Data refreshed daily, snapshot 2026-10-20.