| Claude Sonnet 4.6 | Claude Haiku 4.5 | |
|---|---|---|
| arenaElo | 1358 | 1274 |
| humanEval | 89 | 78.2 |
| sweBench | 60 | 40.6 |
| mmlu | 88.7 | 81.5 |
| gpqa | 59.4 | 48 |
| aime | 80 | 60 |
| Price in/out ($/M) | $3 / $15 | $1 / $5 |
Haiku 4.5 is 3× cheaper and 2× faster than Sonnet 4.6 but gives up ~10 HumanEval points and ~11 GPQA points. Use Haiku for high-volume classification, routing and short chat; escalate to Sonnet whenever quality visibly matters to the end user.
Claude Sonnet 4.6 · Claude Haiku 4.5 · All AI model comparisons