CTF-Archive Diamond
Agents · 2026-05-01
CTF-Archive Diamond is CAISI's difficult 285-challenge selection from the pwn.college CTF collection. Agents run with Inspect's ReAct scaffold, and scores report the percentage of challenges successfully completed.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.5 | 71.0 |
| Claude Mythos Preview | 65.6 |
| Opus 4.8 | 48.4 |
| Opus 4.6 | 46.0 |
| GLM-5.2 | 39.3 |
| GPT-5.4 Mini | 32.0 |
| DeepSeek-V4-Pro | 32.0 |