LLMAPI only
o3
OpenAI
Updated May 21, 2026 · Benchmarks via Artificial Analysis, specs & pricing via OpenRouter · Methodology · Report an error
Specifications
- Type
- LLM
- Access
- API only
- Released
- April 16, 2025
- License
- Proprietary
- Context window
- 200,000 tokens
- Knowledge cutoff
- May 31, 2024
- Output speed
- 50 tok/s
- Latency (TTFT)
- 20s
- Input
- Text, Image
- Output
- Text
- API pricing
- $2.00 in · $8.00 out / 1M tokens
Capabilities
Function callingStructured outputWeb search
Benchmarks
Reasoning
| GPQA Diamond | 87.7 |
| ARC-AGI-2 | 6.5 |
Coding
| Aider Polyglot | 81.3 |
| SWE-bench Verified | 69.1 |
| LiveCodeBench | 80.8 |
Math
| AIME 2024 | 91.6 |
| AIME 2025 | 86.4 |
| FrontierMath | 15.8 |
| MATH-500 | 99.2 |
Agents
| BrowseComp | 49.7 |
| τ²-bench Airline | 64.8 |
| τ²-bench Retail | 80.2 |
General
| Humanity’s Last Exam | 24.3 |
| MMLU-Pro | 85.3 |
Our take
OpenAI’s flagship reasoning model, and the first that could reason over images and call tools mid-thought. It posted large gains on hard math, science, and coding, and folded agentic tool use into the reasoning loop.