Multi-hop Information-Seeking on InfoSeek-Eval ID
82Success Rate (SR)Qwen3-Emb + LRAT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3-Emb + LRATAgent Backbone=GLM-4.7 (358B)2026.03 | 82 | 18.5 | |
| E5-Large + LRATAgent Backbone=GLM-4.7 (358B)2026.03 | 81.7 | 19.5 | |
| Qwen3-Emb + LRATAgent Backbone=MiniMax-M2.1 (229B)2026.03 | 78.3 | 14.7 | |
| E5-Large + LRATAgent Backbone=MiniMax-M2.1 (229B)2026.03 | 75 | 14.8 | |
| E5-LargeAgent Backbone=GLM-4.7 (358B)2026.03 | 73.7 | 24.2 | |
| Qwen3-Emb + LRATAgent Backbone=WebExplore (8B)2026.03 | 68.7 | 19 | |
| Qwen3-Emb + LRATAgent Backbone=Tongyi-DeepResearch (30B)2026.03 | 68 | 20.7 | |
| E5-Large + LRATAgent Backbone=Tongyi-DeepResearch (30B)2026.03 | 68 | 21.5 | |
| Qwen3-EmbAgent Backbone=GLM-4.7 (358B)2026.03 | 67.7 | 27.5 | |
| E5-LargeAgent Backbone=MiniMax-M2.1 (229B)2026.03 | 64 | 18.9 | |
| E5-Large + LRATAgent Backbone=WebExplore (8B)2026.03 | 63.3 | 20.2 | |
| E5-LargeAgent Backbone=WebExplore (8B)2026.03 | 60 | 23.8 | |
| Qwen3-EmbAgent Backbone=MiniMax-M2.1 (229B)2026.03 | 58.7 | 21.4 | |
| E5-LargeAgent Backbone=Tongyi-DeepResearch (30B)2026.03 | 56.7 | 25.1 | |
| Qwen3-Emb + LRATAgent Backbone=AgentCPM-Explore (4B)2026.03 | 55.7 | 34.4 | |
| Qwen3-EmbAgent Backbone=Tongyi-DeepResearch (30B)2026.03 | 52.7 | 26.7 | |
| Qwen3-EmbAgent Backbone=WebExplore (8B)2026.03 | 52 | 24.1 | |
| E5-Large + LRATAgent Backbone=GPT-OSS (120B)2026.03 | 50.7 | 29.7 | |
| E5-Large + LRATAgent Backbone=AgentCPM-Explore (4B)2026.03 | 49.7 | 35.5 | |
| E5-LargeAgent Backbone=AgentCPM-Explore (4B)2026.03 | 47.3 | 38.9 | |
| Qwen3-Emb + LRATAgent Backbone=GPT-OSS (120B)2026.03 | 47 | 30.5 | |
| E5-LargeAgent Backbone=GPT-OSS (120B)2026.03 | 41.7 | 33.9 | |
| Qwen3-EmbAgent Backbone=AgentCPM-Explore (4B)2026.03 | 40.3 | 38 | |
| Qwen3-EmbAgent Backbone=GPT-OSS (120B)2026.03 | 40 | 34.9 |