Question Answering on HotPotQA (EM, F1, Avg Tokens)
68.08F1Hard Routing
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Hard RoutingModel=GPT-4o, Setting=4-source2026.03 | 68.08 | 56.67 | 1,941.2 | |
| Real RouteModel=GPT-4o, Setting=4-source2026.03 | 67.09 | 56.67 | 2,651.3 | |
| Real RouteModel=GPT-4o, Setting=2-source2026.03 | 66.93 | 53.67 | 2,711.5 | |
| Real RouteModel=GPT-4o, Setting=3-source2026.03 | 64.61 | 53 | 2,553.6 | |
| Hard RoutingModel=GPT-4o, Setting=3-source2026.03 | 63.9 | 52.33 | 1,881.6 | |
| Hard RoutingModel=GPT-4o, Setting=2-source2026.03 | 61.7 | 49.3 | 3,926.6 | |
| Hard RoutingModel=GPT-4o-mini, Setting=4-source2026.03 | 60.63 | 48.67 | 2,269.51 | |
| Real RouteModel=GPT-4o-mini, Setting=4-source2026.03 | 60.02 | 49 | 3,183.29 | |
| Real RouteModel=GPT-4o-mini, Setting=2-source2026.03 | 59.61 | 48 | 2,785.6 | |
| Hard RoutingModel=GPT-4o-mini, Setting=3-source2026.03 | 58.57 | 46 | 2,227.7 | |
| INTRAGenerator=T5Gemma2026.05 | 58 | — | — | |
| Real RouteModel=GPT-4o-mini, Setting=3-source2026.03 | 57.53 | 45.67 | 3,051.21 | |
| Hard RoutingModel=GPT-4o-mini, Setting=2-source2026.03 | 56.89 | 46.09 | 2,230.1 | |
| Hybrid RAGGenerator=T5Gemma2026.05 | 54.3 | — | — | |
| Qwen3-Emb-4B + RerankerGenerator=T5Gemma2026.05 | 53.6 | — | — | |
| BGEGenerator=T5Gemma2026.05 | 53 | — | — | |
| MaxSimGenerator=T5Gemma2026.05 | 52.2 | — | — | |
| BM25Generator=T5Gemma2026.05 | 52 | — | — | |
| Qwen3-Emb-4BGenerator=T5Gemma2026.05 | 51.2 | — | — | |
| Qwen3-Emb-0.6BGenerator=T5Gemma2026.05 | 47.7 | — | — | |
| TF-IDFGenerator=T5Gemma2026.05 | 44.5 | — | — |