List your company

Long-horizon Agentic Task Completion on Odysseys

34Average Success Rate (%)MACU
7.4814.36521.2528.135Jun 1, 2026

Evaluation results

MethodLinks
MACUSetup=Multi-Agent, Manager=Opus 3.5, Subagent backbone=Qwen2.5-72B2026.063425.5110.3-52.1
MACUSetup=Multi-Agent, Manager=Qwen2.5-72B, Subagent backbone=Qwen2.5-72B2026.063223.5116.3-46.1
Single-AgentSetup=Single-Agent, Manager=—, Subagent backbone=Qwen2.5-72B2026.068.5—162.4—