Flagship GPT-5.6 model for difficult reasoning, coding, cybersecurity, science, and agentic work.
| Benchmark | Score | Rank |
|---|---|---|
Terminal Agentic terminal coding tasks requiring multi-step execution | 88.8% | #1 / 56 |
AA Coding Independent composite of implementation, terminal use, and real-codebase agent performance | 80% | #1 / 18 |
ARC-AGI Novel reasoning tasks requiring fluid intelligence | 93.8% | #2 / 30 |
MMMU College-level multimodal reasoning across 30+ disciplines | 88.8% | #4 / 57 |
hle | 47.2% | #4 / 76 |
GPQA PhD-level science questions even experts struggle with | 94.1% | #5 / 88 |
AA Intelligence Independent composite of agentic work, coding, scientific reasoning, and general capability | 58.9% | #5 / 18 |
OSWorld Computer use in real desktop environments | 62.6% | #7 / 13 |
MMLU-Pro Harder 10-option successor to MMLU; more reasoning-focused | 89.1% | #13 / 58 |
LiveCodeBench Contamination-free competitive programming (filtered by cutoff date) | 82.6% | #33 / 59 |