First mainline model incorporating GPT-5.3-Codex coding capabilities. Native computer use, 1M context, surpasses human baseline on OSWorld. Codex branch ends here.
| Benchmark | Score | Rank |
|---|---|---|
FrontierMath Unpublished research-level mathematics problems | 47.6% | #1 / 4 |
Terminal Agentic terminal coding tasks requiring multi-step execution | 75.1% | #3 / 48 |
OSWorld Computer use in real desktop environments | 75% | #4 / 9 |
ARC-AGI Novel reasoning tasks requiring fluid intelligence | 73.3% | #4 / 23 |
GPQA PhD-level science questions even experts struggle with | 92.8% | #5 / 64 |
SWE-bench Real-world GitHub issue resolution | 78.2% | #10 / 40 |
MATH Competition-level mathematics problems | 96.7% | #12 / 50 |
HumanEval Coding ability - generating correct Python functions | 94.5% | #13 / 50 |
LiveCodeBenchvals.ai Contamination-free competitive programming (filtered by cutoff date) | 84% | #19 / 40 |
MMMUvals.ai College-level multimodal reasoning across 30+ disciplines | 79.2% | #19 / 39 |
MMLU-Provals.ai Harder 10-option successor to MMLU; more reasoning-focused | 84.6% | #24 / 38 |