Anthropic launches Sonnet 5.5, targeting faster and cheaper tasks
The new model generates responses faster and, Anthropic says, uses fewer tokens for many tasks. Its per-token price remains the same as Sonnet 5’s.
Summary
Anthropic released Claude Sonnet 5.5 on September 28 with a focus on everyday work: faster responses and a lower cost to complete tasks. The company says it generates output more than 30% faster than Sonnet 5 and can cost up to 30% less per task in its testing.
That reduction does not come from a lower per-token price. Through the API, Sonnet 5.5 keeps its predecessor’s rates: $2 per million input tokens and $10 per million output tokens. The claimed saving depends on the model using fewer tokens to reach the same result.
In practice
Anthropic positions Sonnet 5.5 for well-scoped tasks such as fixing software bugs and creating documents, presentations, and spreadsheets. Faster output could make repeated rounds of code or document review feel more responsive. Actual costs will depend on the task, the instructions, and the amount of output produced.
The model is available through Anthropic’s platforms and services including Amazon Web Services, Google Cloud, and Microsoft Azure. Its announced API identifier is `claude-sonnet-5-5`.
What we still don't know
Most of the launch comparisons for speed, performance, and cost per task come from Anthropic’s own evaluations. It remains to be seen how well those results carry over to real workloads with different tasks and quality requirements.
Anthropic also says Sonnet 5.5’s cybersecurity capabilities are comparable to those of Opus 5. It has therefore launched the model with stronger safeguards for higher-risk requests. According to the company, routine software development remains available, though some cybersecurity tasks may fall back to Sonnet 5.
Why it matters
- For teams doing repeated writing, analysis, or coding tasks, speed and token use can matter as much as a model’s peak performance.
- Cost per completed task is a more useful comparison than per-token price alone, provided the quality of the result is measured too.
- The stronger safeguards show that a model aimed at everyday work can still need additional controls as its capabilities grow in sensitive areas.