Web Navigation on WebWalker Easy
58.75Success RateWebWalker
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| WebWalkerLLM Backbone=Qwen-2.5-72B2026.06 | 58.75 | 2.7 | |
| ReflexionLLM Backbone=Qwen-2.5-72B2026.06 | 57.5 | 3.04 | |
| ReflexionLLM Backbone=GPT-4o2026.06 | 56.25 | 2.91 | |
| WebWalkerLLM Backbone=GPT-4o2026.06 | 55 | 2.97 | |
| WebWalkerLLM Backbone=Qwen-Plus2026.06 | 55 | 3.72 | |
| StepGuard (+DDPO+CANR)LLM Backbone=Qwen3B2026.06 | 54.25 | 3.78 | |
| ReActLLM Backbone=GPT-4o2026.06 | 53.75 | 2.53 | |
| ReflexionLLM Backbone=Qwen-Plus2026.06 | 53.75 | 3.66 | |
| StepGuard (+DDPO)LLM Backbone=Qwen3B2026.06 | 52.5 | 2.85 | |
| StepGuard (+CANR)LLM Backbone=Qwen3B2026.06 | 51.81 | 3.86 | |
| ReActLLM Backbone=Qwen-Plus2026.06 | 48.75 | 1.67 | |
| ReActLLM Backbone=Qwen-2.5-32B2026.06 | 47.5 | 2.21 | |
| ReActLLM Backbone=Qwen-2.5-72B2026.06 | 47.5 | 1.68 | |
| StepGuard (Baseline)LLM Backbone=Qwen3B2026.06 | 47.25 | 3.02 | |
| ReflexionLLM Backbone=Qwen-2.5-14B2026.06 | 46.25 | 2.21 | |
| ReflexionLLM Backbone=Qwen-2.5-32B2026.06 | 42.5 | 2.52 | |
| WebWalkerLLM Backbone=Qwen-2.5-7B2026.06 | 41.25 | 3.39 | |
| WebWalkerLLM Backbone=Qwen-2.5-14B2026.06 | 41.25 | 2.42 | |
| WebWalkerLLM Backbone=Qwen-2.5-32B2026.06 | 41.25 | 2.69 | |
| ReActLLM Backbone=Qwen-2.5-7B2026.06 | 37.5 | 3.36 | |
| ReflexionLLM Backbone=Qwen-2.5-7B2026.06 | 37.5 | 4.03 | |
| ReActLLM Backbone=Qwen-2.5-14B2026.06 | 36.25 | 1.86 | |
| WebWalkerLLM Backbone=Qwen3B2026.06 | 36.25 | 3.65 | |
| ReflexionLLM Backbone=Qwen3B2026.06 | 33.75 | 4.21 | |
| ReActLLM Backbone=Qwen3B2026.06 | 32.5 | 3.52 |