| Claude Opus 4.6 | Claude Sonnet 4.6 | |
|---|---|---|
| arenaElo | 1402 | 1358 |
| humanEval | 91.4 | 89 |
| sweBench | 53.4 | 60 |
| mmlu | 90.8 | 88.7 |
| gpqa | 62.1 | 59.4 |
| aime | 84 | 80 |
| Price in/out ($/M) | $15 / $75 | $3 / $15 |
Opus 4.6 still beats Sonnet 4.6 on raw reasoning, but the gap is small and Sonnet costs 5× less. Unless you already validated Opus 4.6 in production or need its 500K context, default to Sonnet 4.6.
Claude Opus 4.6 · Claude Sonnet 4.6 · All AI model comparisons