Web-based Agent Task Completion on WebVoyager
73.9Success Rate (Wolfram)Prompt-Only (Gemini)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Prompt-Only (Gemini)Backbone=Qwen3-30B-A3B-Instruct-25072026.04 | 73.9 | 53.4 | 51.16 | 45.23 | 55.92 | |
| Ours (SFT)Backbone=Seed-OSS-36B-Instruct2026.04 | 71.73 | 51.16 | 53.49 | 47.61 | 56 | |
| Ours (RFT)Backbone=Seed-OSS-36B-Instruct2026.04 | 63.04 | 55.81 | 51.16 | 57.14 | 56.79 | |
| Ours (SFT)Backbone=Qwen3-30B-A3B-Instruct-25072026.04 | 60.87 | 41.86 | 62.79 | 40.48 | 51.5 | |
| Ours (RFT)Backbone=Qwen3-30B-A3B-Instruct-25072026.04 | 58.7 | 48.84 | 67.44 | 54.76 | 57.44 | |
| Prompt-Only (Gemini)Backbone=Seed-OSS-36B-Instruct2026.04 | 58.6 | 53.49 | 62.79 | 52.38 | 56.82 | |
| WithoutBackbone=Qwen3-30B-A3B-Instruct-25072026.04 | 54.3 | 37.2 | 41.86 | 30.95 | 41.08 | |
| Prompt-Only (Base)Backbone=Qwen3-30B-A3B-Instruct-25072026.04 | 54.3 | 32.56 | 42.85 | 33.33 | 40.76 | |
| WithoutBackbone=Seed-OSS-36B-Instruct2026.04 | 54.3 | 48.84 | 23.26 | 33.33 | 39.93 | |
| Prompt-Only (Base)Backbone=Seed-OSS-36B-Instruct2026.04 | 47.82 | 46.34 | 30.23 | 22.5 | 36.72 |