Long-form Question Answering on FAVA MixP (50% polluted)
91.59VeriScore F1@kMIRAGE
Evaluation Results
| Method | Links | |
|---|---|---|
| MIRAGEBackbone=GPT-4o-mini2026.07 | 91.59 | |
| No-RAGBackbone=GPT-4o-mini2026.07 | 81.2 | |
| MIRAGEBackbone=Qwen3-8B2026.07 | 78.7 | |
| DRAGINBackbone=GPT-4o-mini2026.07 | 75.13 | |
| No-RAGBackbone=Qwen3-8B2026.07 | 74.92 | |
| AstuteRAGBackbone=GPT-4o-mini2026.07 | 74.89 | |
| BRIDGEBackbone=GPT-4o-mini2026.07 | 69.46 | |
| RetRobust-NLIBackbone=GPT-4o-mini2026.07 | 69.28 | |
| FLAREBackbone=GPT-4o-mini2026.07 | 68.21 | |
| MADAMRAGBackbone=GPT-4o-mini2026.07 | 64.09 | |
| AstuteRAGBackbone=Qwen3-8B2026.07 | 59.61 | |
| TrustRAGBackbone=GPT-4o-mini2026.07 | 57.19 | |
| RAGBackbone=GPT-4o-mini2026.07 | 55.79 | |
| SelfRAGBackbone=Qwen3-8B2026.07 | 55.37 | |
| BRIDGEBackbone=Qwen3-8B2026.07 | 55.19 | |
| RobustRAGBackbone=Qwen3-8B2026.07 | 54.31 | |
| DRAGINBackbone=Qwen3-8B2026.07 | 53.68 | |
| MADAMRAGBackbone=Qwen3-8B2026.07 | 52.38 | |
| SelfRAGBackbone=GPT-4o-mini2026.07 | 51.86 | |
| RARGBackbone=GPT-4o-mini2026.07 | 51.63 | |
| RetRobust-NLIBackbone=Qwen3-8B2026.07 | 50.17 | |
| TrustRAGBackbone=Qwen3-8B2026.07 | 49.23 | |
| FLAREBackbone=Qwen3-8B2026.07 | 48.64 | |
| RAGBackbone=Qwen3-8B2026.07 | 47.62 | |
| RobustRAGBackbone=GPT-4o-mini2026.07 | 46.77 | |
| RARGBackbone=Qwen3-8B2026.07 | 45.74 |