Anthropic announced Claude Sonnet 5.5 on Sept. 28, 2026, as the second model in the Claude 5.5 family. The company positions it for well-scoped everyday work, including coding, bug fixes, documents, spreadsheets and interface design, and says it generates output more than 30% faster than Sonnet 5.
What changes from Sonnet 5
Anthropic says Sonnet 5.5 can cost up to 30% less per task than Sonnet 5 for most work, attributing the potential savings to using fewer tokens and working more efficiently. That is a comparison with Sonnet 5—not Opus 5.5—and the listed per-token rates are unchanged:
- Input: $2 per million tokens
- Output: $10 per million tokens
- Cache reads: $0.20 per million tokens
- Cache writes: $2.50 per million tokens
A token is a unit of text processed or generated by a model. The rates are API prices in US dollars; the cost of a completed task also depends on how many tokens it uses. For example, Base44 reported an average of 3.6 iterations per build across 118 app builds with Sonnet 5.5, compared with 7.7 using Opus 5. That is one company’s evaluation, not a general cost estimate.
How Sonnet 5.5 compares with Opus 5.5
Anthropic’s results vary by benchmark. CursorBench 4.0, which uses ambiguous, multi-file tasks, puts Opus 5.5 slightly ahead. GDPval-AA scores are also close; Anthropic says a structured-output bug affected Sonnet’s prerelease run and was fixed before launch.
| Evaluation | Claude Sonnet 5.5 | Claude Opus 5.5 |
| CursorBench 4.0 | 55.5% | 57.8% |
| GDPval-AA v2.1 (Elo-style score) | 1,844 | 1,846 |
Anthropic also reports 70.6% for Sonnet 5.5 and 66.4% for Opus 5.5 on Terminal-Bench 4.0. Its benchmark note specifies Xhigh effort for Opus, so those headline scores do not compare the models at the same effort setting. They are results on that test, not a verdict across every coding task.
The cost of a completed task can also differ from the per-token rate. In one ten-evaluation index workload at Max effort, estimated costs were $7.60 per Sonnet 5.5 task and $5.98 per Opus 5.5 task. A separate index using a different task mix estimated $20.80 for Sonnet 5.5 and $32.77 for Opus 5.5. Those figures apply to their respective workloads, not to a general API tariff.
Which work does Anthropic position Sonnet for?
Anthropic positions Sonnet 5.5 for defined, everyday tasks and Opus 5.5 for complex work that calls for sustained judgment. The default effort setting is Medium in the Claude app and Claude Code, and High on the Claude Platform—a practical distinction when comparing results from different settings.
The model also has safeguards for cybersecurity tasks. Anthropic says higher-risk requests visibly fall back to Sonnet 5, while routine software development remains available.
Availability and the next Claude 5.5 model
Anthropic said Sonnet 5.5 was available on all platforms, including the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure. In the Sept. 28 announcement, the company said Haiku 5.5 would join the family in the coming weeks, without giving a specific release date.
