Web Agent Navigation on WorkArena L2 147-task (test)
40Success RateGemini 3.1 Pro
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini 3.1 ProAgent harness=This work2026.04 | 40 | |
| Claude 3.5 SonnetAgent harness=BrowserGym ecosystem2026.04 | 38.8 | |
| Qwen3.5-27BParams=27B, Agent harness=This work2026.04 | 18.9 | |
| o1-miniAgent harness=BrowserGym ecosystem2026.04 | 14.3 | |
| A3-Qwen3.5-9BParams=9B, Agent harness=This work2026.04 | 9.7 | |
| Llama 3.1 405BParams=405B, Agent harness=BrowserGym ecosystem2026.04 | 8.9 | |
| GPT-4oAgent harness=BrowserGym ecosystem2026.04 | 7.5 | |
| Llama 3.1 70BParams=70B, Agent harness=BrowserGym ecosystem2026.04 | 3.4 | |
| Qwen3.5-9B (base)Params=9B, Agent harness=This work2026.04 | 2.2 | |
| GPT-4o-miniAgent harness=BrowserGym ecosystem2026.04 | 2 |