Multi-Hop Question Answering on HotpotQA (Accuracy)
44.2AccuracySkillOrchestra+
Evaluation Results
| Method | Links | |
|---|---|---|
| SkillOrchestra+Routing Strategy=Skill-based Routing, Orchestrator Configuration=Best/Multiple2026.02 | 44.2 | |
| HCAPO (Ours)Type=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 42.1 | |
| GiGPOType=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 41.6 | |
| SkillOrchestraRouting Strategy=Skill-based Routing, Orchestrator Configuration=Single2026.02 | 39 | |
| HCAPO (Ours)Type=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 38.6 | |
| StepSearchType=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 38.6 | |
| Search-R1Type=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 37 | |
| GiGPOType=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 36.9 | |
| Router-R1Routing Strategy=RL-based Routing2026.02 | 35.2 | |
| ZeroSearchType=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 34.6 | |
| StepSearchType=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 34.5 | |
| Search-R1Type=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 32.4 | |
| Largest LLMRouting Strategy=Heuristic & Discriminative Routing2026.02 | 27.8 | |
| ZeroSearchType=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 27.4 | |
| Prompt LLMRouting Strategy=Heuristic & Discriminative Routing2026.02 | 26.8 | |
| RouterDCRouting Strategy=Heuristic & Discriminative Routing2026.02 | 24.4 | |
| R1-InstructType=RL Training, Base Model=Qwen2.5-3B-Instruct2026.03 | 23.7 | |
| Search-R1Routing Strategy=No Routing2026.02 | 23.6 | |
| GraphRouterRouting Strategy=Heuristic & Discriminative Routing2026.02 | 23.4 | |
| FrugalGPTRouting Strategy=Heuristic & Discriminative Routing2026.02 | 23.4 | |
| KNN RouterRouting Strategy=Heuristic & Discriminative Routing2026.02 | 22.4 | |
| RAGRouting Strategy=No Routing2026.02 | 21.6 | |
| BERT RouterRouting Strategy=Heuristic & Discriminative Routing2026.02 | 21.6 | |
| R1-InstructType=RL Training, Base Model=Qwen2.5-7B-Instruct2026.03 | 20.8 | |
| Prompt LLM+Routing Strategy=Heuristic & Discriminative Routing, Turn Protocol=multi turn2026.02 | 20.6 | |
| SFTRouting Strategy=No Routing2026.02 | 19.8 | |
| MLP RouterRouting Strategy=Heuristic & Discriminative Routing2026.02 | 19.8 | |
| CoTRouting Strategy=No Routing2026.02 | 16.8 | |
| KNN Router+Routing Strategy=Heuristic & Discriminative Routing, Turn Protocol=multi turn2026.02 | 15.4 | |
| VanillaRouting Strategy=No Routing2026.02 | 14 |