Web Navigation on WebWalker Medium
51.43Success Rate (SR)Reflexion
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ReflexionLLM Backbone=GPT-4o2026.06 | 51.43 | 3.88 | |
| WebWalkerLLM Backbone=GPT-4o2026.06 | 50 | 3.43 | |
| ReActLLM Backbone=Qwen-Plus2026.06 | 48.57 | 2.69 | |
| WebWalkerLLM Backbone=Qwen-2.5-72B2026.06 | 48.57 | 3.07 | |
| StepGuard (+DDPO+CANR)LLM Backbone=Qwen3B2026.06 | 47.99 | 4.2 | |
| WebWalkerLLM Backbone=Qwen-Plus2026.06 | 47.14 | 3.19 | |
| StepGuard (+DDPO)LLM Backbone=Qwen3B2026.06 | 45.71 | 3.28 | |
| ReActLLM Backbone=GPT-4o2026.06 | 45 | 3.34 | |
| ReflexionLLM Backbone=Qwen-2.5-72B2026.06 | 44.29 | 3.88 | |
| StepGuard (+CANR)LLM Backbone=Qwen3B2026.06 | 44.2 | 4.49 | |
| WebWalkerLLM Backbone=Qwen-2.5-14B2026.06 | 41.43 | 3.46 | |
| StepGuard (Baseline)LLM Backbone=Qwen3B2026.06 | 40.43 | 3.55 | |
| ReflexionLLM Backbone=Qwen-Plus2026.06 | 40 | 3.79 | |
| ReActLLM Backbone=Qwen-2.5-72B2026.06 | 38.57 | 2.79 | |
| ReActLLM Backbone=Qwen-2.5-32B2026.06 | 35.71 | 2.3 | |
| ReflexionLLM Backbone=Qwen-2.5-14B2026.06 | 34.29 | 2.83 | |
| WebWalkerLLM Backbone=Qwen-2.5-32B2026.06 | 34.29 | 3.09 | |
| ReflexionLLM Backbone=Qwen-2.5-32B2026.06 | 32.86 | 2.65 | |
| ReActLLM Backbone=Qwen-2.5-14B2026.06 | 32.14 | 2.75 | |
| ReflexionLLM Backbone=Qwen-2.5-7B2026.06 | 25 | 3.48 | |
| WebWalkerLLM Backbone=Qwen-2.5-7B2026.06 | 24.71 | 3.86 | |
| ReflexionLLM Backbone=Qwen3B2026.06 | 21.43 | 3.65 | |
| WebWalkerLLM Backbone=Qwen3B2026.06 | 20 | 4.1 | |
| ReActLLM Backbone=Qwen-2.5-7B2026.06 | 18.57 | 4.88 | |
| ReActLLM Backbone=Qwen3B2026.06 | 15.71 | 5.12 |