| Claude Opus 4.7 | Claude Haiku 4.5 | |
|---|---|---|
| arenaElo | 1420 | 1274 |
| humanEval | 93.1 | 78.2 |
| sweBench | 64.3 | 40.6 |
| mmlu | 92.3 | 81.5 |
| gpqa | 65.2 | 48 |
| aime | 88 | 60 |
| Price in/out ($/M) | $15 / $75 | $1 / $5 |
Opus 4.7 is 15× more expensive than Haiku 4.5 and delivers the quality to match on reasoning and coding. Haiku 4.5 is the right default for volume and latency-bound UX. A two-model strategy (Haiku for routing, Opus for depth) beats picking one.
Claude Opus 4.7 · Claude Haiku 4.5 · All AI model comparisons