Long-horizon Agentic Task Completion on Odysseys
34Average Success Rate (%)MACU
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MACUSetup=Multi-Agent, Manager=Opus 3.5, Subagent backbone=Qwen2.5-72B2026.06 | 34 | 25.5 | 110.3 | -52.1 | |
| MACUSetup=Multi-Agent, Manager=Qwen2.5-72B, Subagent backbone=Qwen2.5-72B2026.06 | 32 | 23.5 | 116.3 | -46.1 | |
| Single-AgentSetup=Single-Agent, Manager=—, Subagent backbone=Qwen2.5-72B2026.06 | 8.5 | — | 162.4 | — |