ASI-Bench: At the Dawn of Artificial Superintelligence
Published in arXiv preprint arXiv:2608.17271, 2026
ASI-Bench evaluates frontier AI systems on research-grade scientific work at a scale beyond existing expert benchmarks. Its 60 project-level tasks span 11 scientific domains, with B1-B4 guidance-withdrawal control to measure how far agents progress as scaffolding is removed.
Links
- Paper (arXiv): https://arxiv.org/abs/2608.17271
- Website: https://asibench.apexin.ai/
- Code: https://github.com/apexin-ai/ASI-Bench
Recommended citation: Zhou, J., et al. (including Wu, K.) (2026). "ASI-Bench: At the Dawn of Artificial Superintelligence." arXiv preprint arXiv:2608.17271.
Download Paper