| GPQA Diamond(OpenAI-reported maximum at any effort setting) | 96% |
| HLE(Artificial Analysis, max effort) | 54.7% |
| Terminal-Bench 2.1(Artificial Analysis, high effort) | 89.9% |
| Terminal-Bench 4.0(OpenAI-reported maximum; Anthropic reports the same 57.9 at high effort) | 57.9% |