Question Answering on GAIA (test)
41.7Unreliability RateKbSD
Evaluation Results
| Method | Links | |
|---|---|---|
| KbSDBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 41.7 | |
| KbSDBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 58.1 | |
| KbPOBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 71.9 | |
| KbPOBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 77.2 | |
| Search-R1Backbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 89.92 | |
| ARPOBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 91.44 | |
| TCRAGBackbone=Qwen2.5-7B, Paradigm=Agentic RAG2026.06 | 91.79 | |
| ARPOBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 92.29 | |
| COTBackbone=Qwen2.5-3B, Paradigm=No RAG2026.06 | 92.98 | |
| Mem1Backbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 93.5 | |
| ReACTBackbone=Qwen2.5-7B, Paradigm=Agentic RAG2026.06 | 93.63 | |
| COTBackbone=Qwen2.5-7B, Paradigm=No RAG2026.06 | 93.83 | |
| BaseBackbone=Qwen2.5-3B, Paradigm=No RAG2026.06 | 94.3 | |
| TCRAGBackbone=Qwen2.5-3B, Paradigm=Agentic RAG2026.06 | 94.47 | |
| IRCOTBackbone=Qwen2.5-7B, Paradigm=Agentic RAG2026.06 | 94.5 | |
| FS-RAGBackbone=Qwen2.5-7B, Paradigm=Naive RAG2026.06 | 94.98 | |
| AEPOBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 94.99 | |
| ReACTBackbone=Qwen2.5-3B, Paradigm=Agentic RAG2026.06 | 95.22 | |
| Search-R1Backbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 95.32 | |
| FL-RAGBackbone=Qwen2.5-7B, Paradigm=Naive RAG2026.06 | 95.4 | |
| ReSearchBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 95.52 | |
| ReSearchBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 95.57 | |
| FS-RAGBackbone=Qwen2.5-3B, Paradigm=Naive RAG2026.06 | 95.7 | |
| BaseBackbone=Qwen2.5-7B, Paradigm=No RAG2026.06 | 95.71 | |
| IRCOTBackbone=Qwen2.5-3B, Paradigm=Agentic RAG2026.06 | 97.23 | |
| AEPOBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 97.35 | |
| Mem1Backbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 97.53 | |
| FL-RAGBackbone=Qwen2.5-3B, Paradigm=Naive RAG2026.06 | 98.42 |