o3
OpenAI·
LLMsreasoningcloud
Overview
Full reasoning model with significant improvements over o1.
Capabilities and innovations
Advanced reasoningTool useAgentic codingImproved Reasoning ChainTool IntegrationMulti-step Planning
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 82.5% | unsourced |
| Humanity’s Last Exam (no tools) | 28.4% | unsourced |
| SWE-bench Verified | 69.1% | unsourced |
| MMLU | 91.6% | unsourced |
| MMLU-Pro | 85.59% | unsourced |
| ARC-AGI-1 | 75.7% | independent ARC-AGI-1 semi-private eval, verified by ARC Prize at the public $10k compute limit; a high-compute (172x) configuration reached 87.5%. |
- SWE-bench Pro: benchmark did not exist at release
- Terminal-Bench 2.x: benchmark did not exist at release
- MMMU-Pro: benchmark did not exist at release
Links
More from OpenAI
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.