Web Navigation on WebWalker 100 tasks (test)
0.125Success Rate (Easy)Vanilla GRPO
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Vanilla GRPOBackbone=Qwen3-4b2026.05 | 0.125 | 0.2206 | 0.25 | 22 | |
| Skill-R1Mode=GRPO, Backbone=Qwen3-4B2026.05 | 0.125 | 0.2941 | 0.2083 | 26 | |
| GPT-4o-miniSkills=no skills2026.05 | 0 | 0.0147 | 0.0417 | 2 | |
| Skill-R1Mode=Inference, Backbone=Qwen3-4B2026.05 | 0 | 0.2059 | 0.2083 | 19 |