OpenAI Capture-the-Flag challenges - internal release score
Code
Expanded hard Capture-the-Flag challenge set in OpenAI's model releases. Exact cohort and version comparability are undisclosed; diagnostic rather than a standard public CTF accuracy.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 96.7 |
| GPT-5.6 Terra | 91.8 |
| GPT-5.5 | 88.1 |
| GPT-5.6 Luna | 85.2 |