ExploitBench - Internal Port (June-August 2026)
Code
OpenAI's internal evaluation of whether models develop working exploits for vulnerabilities disclosed in June-August 2026, built to reduce contamination from historical ExploitBench examples; reported as the arbitrary-code-execution success rate. OpenAI does not publish its task count, harness or trial count. Its GPT-6 Astra figure (31.5%) differs from the 39% Astra launch-post figure on ExploitBench June-August 2026, so the two are kept apart.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 31.5 |
| GPT-6.1 Terra | 24.5 |
| GPT-6.1 Sol | 21.5 |
| GPT-6 Sol | 5.5 |
| GPT-5.6 Sol | 3.5 |