| GPT-4o | GPT-5 mini | |
|---|---|---|
| arenaElo | 1345 | 1288 |
| humanEval | 90.2 | 81 |
| sweBench | 33 | 50 |
| mmlu | 88.7 | 82.4 |
| gpqa | 53.6 | 50.1 |
| aime | 42 | 78 |
| Price in/out ($/M) | $2.5 / $10 | $0.25 / $2 |
GPT-5 mini is cheaper, smarter on most benchmarks and has 3× the context of GPT-4o. GPT-4o only wins when you need the Realtime voice API. New projects should default to GPT-5 mini.