Multi-hop Question Answering on HotpotQA (Expected utility)
88.4Expected UtilityFlowCompile
Evaluation Results
| Method | Links | |
|---|---|---|
| FlowCompileWorkflow Category=Main Method2026.05 | 88.4 | |
| Pref-Aware MaASWorkflow Category=Workflow with Router2026.05 | 78.7 | |
| MaAS (Qwen3-4B)Workflow Category=Workflow with Router2026.05 | 74.2 | |
| KNN RouterWorkflow Category=Workflow with Router2026.05 | 74.1 | |
| Pref-Aware RouterWorkflow Category=Workflow with Router2026.05 | 69.3 | |
| Qwen3-4BWorkflow Category=Fixed Workflow2026.05 | 68.3 | |
| Qwen3-32BWorkflow Category=Single Model2026.05 | 67.4 | |
| MaAS (Qwen3-8B)Workflow Category=Workflow with Router2026.05 | 65.2 | |
| QwQ-32BWorkflow Category=Single Model2026.05 | 62.6 | |
| Qwen3-8BWorkflow Category=Fixed Workflow2026.05 | 58.1 |