UI-TARS
Pioneering Automated GUI Interaction with Native Agents
UI-TARS scores 4.4/10 (Emerging) in Browser & Computer Use.
Language Python License Apache-2.0 Stars 11.1k★ Type model Self-host Yes Repo bytedance/UI-TARS Homepage https://github.com/bytedance/UI-TARS
Quality
Emerging 4.4
Adoption 51
Activity 11
Maturity 78
Community 50
Capability 25
Show the math
Overall 4.4/10
value = 4.4/10
score = value
Adoption 5.6/10
value = 51.2/100
score = 1 + 9 * value / 100
Activity 2/10
value = 11/100
score = 1 + 9 * value / 100
Maturity 8/10
value = 77.9/100
score = 1 + 9 * value / 100
Community 5.5/10
value = 49.5/100
score = 1 + 9 * value / 100
Capability 3.2/10
value = 24.6/100
score = 1 + 9 * value / 100
Key metrics
11.1k★ GitHub stars
0/90d Recent commits
0 devs/90d Recent contributors
OSWorld 24.6 OSWorld success rate (UI-TARS-72B-DPO, 50 steps) UI-TARS-72B-DPO scores 24.6 on OSWorld at 50 steps (22.7 at 15 steps), outperforming Claude (22.0). Source: official UI-TARS paper https://arxiv.org/abs/2501.12326 (also huggingface.co/papers/2501.12326). Note: a low absolute number reflects OSWorld difficulty (human ~72.4%); higher is still better here.
1.5y Project age
5mo ago Last commit
Gotchas
No gotchas documented yet. Contribute one if you know a constraint we missed.
Labels
- Self-hosted
Self-hosted
You deploy and operate it yourself; there is no hosted option here.
gui vision model grounding computer-use