Claude Sonnet 5.5: what changed in speed and task costs

On September 28, Anthropic introduced Claude Sonnet 5.5 for everyday document and coding work. The company reports output generation over 30% faster than Sonnet 5 and costs up to 30% lower for most tasks. These are the developer’s findings, not our measurements.

What “cheaper” means

Base Claude API rates are $2 per million input tokens and $10 per million output tokens; cache reads cost $0.20 per million. Tokens measure the volume processed by the model. The announcement says these rates are unchanged from Sonnet 5: the expected saving comes from using fewer tokens to complete a task.

An illustrative standard API calculation without caching: 100,000 input tokens and 10,000 output tokens cost $0.30 at those two rates. That covers only the stated volume; additional tools and retries can change the total. Evaluate your own bill against completed tasks that produced an acceptable result.

A Claude subscription and API access are billed separately. Token pricing therefore does not describe the monthly cost of Pro or Max. If you only use the chat interface, check the model and usage limits available on your plan.

Choosing between speed and complexity

Anthropic’s explanation of effort levels recommends starting at the model’s default. For a simple recurring task, lower the setting and compare quality; for a difficult one, raise it. Identical effort names on different models do not imply identical amounts of work.

Our practical criterion is to assess the finished document, beyond how quickly text appears. For example, ask for several emails to be turned into a list of agreements, owners and deadlines. Check for missing tasks and invented dates. If a fast answer requires lengthy corrections, the time saving disappears.

Anthropic positions Sonnet as a faster complement to Opus 5.5, which it favors for more complex, open-ended work. Our editorial advice is to start with a task that has a clear outcome and compare models against the same requirements.

What developers should know before switching

The Claude API identifier is claude-sonnet-5-5. The migration guide lists breaking changes: forced tool selection with tool_choice set to any or tool returns an error. Use between_tools at high effort or lower to turn off up-front thinking. These points concern API integrations rather than an ordinary chat request.

For an application, first check a representative workflow: answer accuracy, tool calls, error handling and actual cost. For chat use, begin with a small task whose result is easy to verify. A new version number does not replace that check.