| GPT-5 mini | Gemini 3 Flash | |
|---|---|---|
| arenaElo | 1288 | 1298 |
| humanEval | 81 | 79.4 |
| sweBench | 50 | 48 |
| mmlu | 82.4 | 83.2 |
| gpqa | 50.1 | 51.8 |
| aime | 78 | 76 |
| Price in/out ($/M) | $0.25 / $2 | $0.3 / $2.5 |
GPT-5 mini is marginally smarter at $0.25/M input; Gemini 3 Flash costs slightly more but brings a 1M context window. For chat UX at scale, GPT-5 mini. For cheap long-doc jobs, Gemini 3 Flash.