Web-based Agent Task Completion on WebWalker
53.5Success Rate (Config)Prompt-Only (Gemini)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Prompt-Only (Gemini)Backbone=Seed-OSS-36B-Instruct2026.04 | 53.5 | 24.37 | 23.96 | 23.48 | 31.33 | |
| Ours (RFT)Backbone=Seed-OSS-36B-Instruct2026.04 | 45.07 | 34.29 | 38.41 | 32.22 | 37.5 | |
| Ours (SFT)Backbone=Qwen3-30B-A3B-Instruct-25072026.04 | 45.05 | 37.35 | 37.98 | 32.31 | 38.17 | |
| Ours (RFT)Backbone=Qwen3-30B-A3B-Instruct-25072026.04 | 43.14 | 42.47 | 42.16 | 35.86 | 40.91 | |
| Prompt-Only (Gemini)Backbone=Qwen3-30B-A3B-Instruct-25072026.04 | 35.59 | 27.87 | 31.36 | 24.56 | 29.85 | |
| Ours (SFT)Backbone=Seed-OSS-36B-Instruct2026.04 | 35.48 | 24.1 | 26.8 | 27.9 | 28.57 | |
| WithoutBackbone=Qwen3-30B-A3B-Instruct-25072026.04 | 24.28 | 23.65 | 22.3 | 17.93 | 22.04 | |
| Prompt-Only (Base)Backbone=Qwen3-30B-A3B-Instruct-25072026.04 | 21.37 | 20.91 | 17.42 | 18.29 | 19.5 | |
| Prompt-Only (Base)Backbone=Seed-OSS-36B-Instruct2026.04 | 20.51 | 12.2 | 17.42 | 16.46 | 16.65 | |
| WithoutBackbone=Seed-OSS-36B-Instruct2026.04 | 19.37 | 10.75 | 21.8 | 13.11 | 16.26 |