OpenAI o4-mini
Last updated: April 22, 2026
Specs
| Vendor | OpenAI |
|---|
| Released | February 12, 2026 |
|---|
| Context window | 200K tokens |
|---|
| Input | $1.1 / M tokens |
|---|
| Output | $4.4 / M tokens |
|---|
Benchmarks
| mmlu | 85.4 |
|---|
| gpqa | 74.2 |
|---|
| humanEval | 88.6 |
|---|
| arenaElo | 1362 |
|---|
| sweBench | 60 |
|---|
| sweBenchPro | 50 |
|---|
| liveCodeBench | 78 |
|---|
| aiderPolyglot | 72 |
|---|
| terminalBench | 40 |
|---|
| mmluPro | 80 |
|---|
| hle | 18 |
|---|
| aime | 92 |
|---|
| math500 | 96 |
|---|
| mmmu | 74 |
|---|
Strengths
- Reasoning quality close to o3 at a fraction of the price
- Great on math, coding and analytical tasks
- Fast enough for interactive agent loops
Weaknesses
- Still slower than GPT-5 mini per request
- Smaller multimodal breadth than GPT-5 family
- Thinking tokens can surprise you on cost
Best for
- Cost-sensitive reasoning agents
- Classroom + tutoring math assistants
- Code review at scale
Compare with other AI models