OpenAI o3
Last updated: August 10, 2026
Specs
| Vendor | OpenAI |
|---|
| Released | April 16, 2025 |
|---|
| Context window | 200K tokens |
|---|
| Input | $2 / M tokens |
|---|
| Output | $8 / M tokens |
|---|
Benchmarks
| mmlu | 90.2 |
|---|
| gpqa | 83.3 |
|---|
| humanEval | 92.4 |
|---|
| arenaElo | 1418 |
|---|
| sweBench | 69.1 |
|---|
| sweBenchPro | 56 |
|---|
| liveCodeBench | 80 |
|---|
| aiderPolyglot | 79 |
|---|
| terminalBench | 48 |
|---|
| mmluPro | 85 |
|---|
| hle | 20.3 |
|---|
| aime | 91.6 |
|---|
| math500 | 97 |
|---|
| mmmu | 78 |
|---|
Strengths
- Best-in-class reasoning (GPQA 83%)
- Deliberate thinking mode for hard problems
- Excellent coding on competitive benchmarks
Weaknesses
- Slow. Thinks for seconds to minutes before answering
- Expensive per interaction due to reasoning tokens
- Not suitable for realtime UX
Best for
- Scientific and mathematical problem solving
- Code review and complex debugging
- Research assistants for experts
Compare with other AI models