Multi-hop Question Answering on MuSiQue (F1, Token Usage, Compression)
68.98F1 ScoreOrgAgent
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OrgAgentBackbone=GPT-5mini, Organization=Hierarchical (AUTO)2026.04 | 68.98 | 11,408 | 37.11 | 59.94 | |
| OrgAgentBackbone=GPT-OSS-120B, Organization=Hierarchical (AUTO)2026.04 | 57.58 | 12,046 | 18.97 | 52.22 | |
| GPT-5miniOrganization=Baseline2026.04 | 51.28 | 2,778 | — | — | |
| GPT-5miniOrganization=Flat2026.04 | 50.31 | 28,479 | — | — | |
| GPT-OSS-120BOrganization=Flat2026.04 | 48.4 | 25,209 | — | — | |
| GPT-OSS-120BOrganization=Baseline2026.04 | 37.98 | 2,687 | — | — | |
| OrgAgentBackbone=LLaMA-3.1-8B, Organization=Hierarchical (AUTO)2026.04 | 32.59 | 12,322 | 123.99 | 76.04 | |
| LLaMA-3.1-8BOrganization=Flat2026.04 | 14.55 | 51,425 | — | — | |
| LLaMA-3.1-8BOrganization=Baseline2026.04 | 11.52 | 2,370 | — | — |