BrowserGym MiniWoB: leaderboard
BrowserGym evaluation on MiniWoB++: 125 simple web interaction tasks (form filling, clicking, navigation) using a standardized GenericAgent framework.
Metric: Success Rate (%). Source: miniwob.farama.org. Source: huggingface.co. Saturation forecast: Around January 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 71.5 | #98 |
| 2 | GPT-5 Mini | 71 | #186 |
| 3 | Claude Sonnet 4 | 70.7 | #199 |
| 4 | Claude 3.5 Sonnet | 69.8 | #350 |
| 5 | GPT-OSS-120B | 66.4 | #341 |
| 6 | GPT-5 Nano | 64.8 | #425 |
| 7 | Llama 3.1 405B | 64.6 | #466 |
| 8 | GPT-OSS-20B | 64 | #518 |
| 9 | GPT-4o | 63.8 | #343 |
| 10 | Llama 3.1 70B | 57.6 | #592 |
| 11 | GPT-4o Mini | 56.6 | #599 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=browsergym-miniwob · How It Works · Data refreshed daily, snapshot 2026-10-20.