| Gemini 3 Pro | Claude Sonnet 4.6 | |
|---|---|---|
| arenaElo | 1405 | 1358 |
| humanEval | 87.3 | 89 |
| sweBench | 63 | 60 |
| mmlu | 91.1 | 88.7 |
| gpqa | 62.5 | 59.4 |
| aime | 90 | 80 |
| Price in/out ($/M) | $7 / $21 | $3 / $15 |
Gemini 3 Pro is stronger on benchmarks and has 10× the context, but costs 2× Sonnet 4.6. For multimodal long-doc work, Gemini Pro. For text chat and coding copilots at scale, Sonnet 4.6.