Claude Opus 5.5
Anthropic's current flagship. At max effort it leads the 1 Oct 2026 Intelligence Index at 58. At high effort it is 54, for $1.82 a measured task.
Every published effort
| Setting | Index | Per task | Tokens / sec |
|---|---|---|---|
| Max | 58 | $5.98 | 92 |
| Extra high | 56 | $3.46 | 79 |
| High | 54 | $1.82 | 74 |
| Medium | 51 | $1.34 | 72 |
| Low | 42 | $0.55 | 72 |
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.
Specs
- API id
- claude-opus-5-5
- Released
- 22 Sep 2026
- Context
- 1M
- Max output
- 128k
- Input / 1M
- $4.00
- Output / 1M
- $20.00
- Cache read / 1M
- $0.20
- Knowledge
- Jun 2026
- Inputs
- Text, Image
- Weights
- Closed
Cache reads are 5% of input, cheaper than the usual 10%. Batch is half. Fast mode is $8 / $40. US-only inference is 1.1×. The 1M window has no long-context surcharge.
Use it when
Long coding sessions and knowledge work, when you want the highest generally available score on this index. High effort is the setting to buy. Max effort adds a few points and about triples the measured task cost.
Leave it when
High-volume calls, and any job GPT-6.1 Sol or MiMo-V2.6-Pro already does well enough. Opus is the quality pick, not the thrifty one.
Other independent scores
- Terminal-Bench 4.0: 59.6%. Max effort. A later write-up rounds Opus and Astra to 60%. Artificial Analysis, Opus 5.5 article, 22 Sep 2026.
- AutomationBench-AA: 70%. Max effort, compared in the Gemini 4 Argon article with Sonnet 5.5 at 71%. Artificial Analysis, 30 Sep 2026.