Claude Sonnet 5.5
The fast Claude. Two points behind Opus at max effort on the index, and more expensive per measured task because it spends more output tokens.
47Index · High
$1.08Per measured task
43.5 pts/$Points / $
94Tokens / sec
Every published effort
| Setting | Index | Per task | Tokens / sec |
|---|---|---|---|
| Max | 56 | $7.62 | 139 |
| Extra high | 52 | $2.74 | 103 |
| High | 47 | $1.08 | 94 |
| Medium | 41 | $0.59 | 92 |
What this workload costs
List price for one call. Reasoning tokens count as output. A cache hit uses the lab’s cache rate when one is published.
Claude Sonnet 5.5
Cache appliedSpecs
- API id
- claude-sonnet-5-5
- Released
- 28 Sep 2026
- Context
- 1M
- Max output
- 128k
- Input / 1M
- $2.00
- Output / 1M
- $10.00
- Cache read / 1M
- $0.20
- Knowledge
- —
- Inputs
- Text, Image
- Weights
- Closed
Same sticker as Sonnet 5. Cache reads stay at $0.20. Max effort is expensive per task because the model writes a lot of tokens, not because the token price changed.
Use it when
Everyday Claude work, edits, and terminal tasks. Extra-high is the compromise. Max effort is a frontier coding setting, and it was one of the most expensive measured tasks in this snapshot.
Leave it when
Leaving max effort on for routine traffic. The token price looks like Sol. The task cost does not.
Other independent scores
- Terminal-Bench 4.0: 64%. Max effort. Ahead of Opus 5.5 and GPT-6 Astra in that write-up. Artificial Analysis, Sonnet 5.5 article, 28 Sep 2026.
- AutomationBench-AA: 71%. Max effort. Artificial Analysis, 30 Sep 2026.