CyberGym-E2E S2 (615 tasks)
Agents · 2026-06-03
CyberGym-E2E cumulative S2 success rate on the earlier 615-task set. A task passes when the agent first generates a crashing proof of concept and then patches the crash.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3 Pro Preview | 23.6 |
| GPT-5.2 | 22.0 |
| Opus 4.5 | 21.9 |
| Sonnet 4.5 | 12.1 |