| Claude Sonnet 4.6 | Claude Sonnet 4.5 | |
|---|---|---|
| arenaElo | 1358 | 1332 |
| humanEval | 89 | 86.1 |
| sweBench | 60 | 58 |
| mmlu | 88.7 | 86.9 |
| gpqa | 59.4 | 56.2 |
| aime | 80 | 76 |
| Price in/out ($/M) | $3 / $15 | $3 / $15 |
Sonnet 4.6 adds ~3 HumanEval points and ~3 GPQA points over 4.5 with the same pricing and context window. Tool use is sharper and agent drift is lower. If you ship on Sonnet, 4.6 is a free quality upgrade: migrate.
Claude Sonnet 4.6 · Claude Sonnet 4.5 · All AI model comparisons