Anthropic released Claude Sonnet 5.5 on September 28, 2026, positioning it as the faster, lower-cost everyday model in the Claude 5.5 family. It is not meant to replace Opus 5.5 for complex judgment. Instead, it targets well-scoped coding, bug fixes, documents, slides, and spreadsheets where shorter waits and lower task cost matter.
Anthropic says Sonnet 5.5 generates output more than 30% faster than Sonnet 5. The token price remains $2 per million input tokens, $10 per million output tokens, and $0.20 per million cache reads. The company’s claim is that the model usually needs fewer tokens for the same work, bringing task cost down by up to 30%; the actual result depends on effort level, tool calls, and task length.
In Anthropic’s published evaluations, Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0 compared with 10.3% for Sonnet 5, with gains also reported on FrontierCode, CursorBench, and AA-Briefcase. These are vendor and partner evaluation results, not a guarantee for every repository, team workflow, or production agent. Anthropic also says Opus 5.5 remains stronger on complex, open-ended work that needs sustained judgment.
The model puts effort level at the center of the speed and cost trade-off. Claude apps default to Medium effort while the Claude Platform defaults to High effort. Lower settings suit fast, well-specified work; higher settings spend more time checking the result. Model routing therefore needs to choose not only a model, but also the depth, risk, and cost allowed for the task.
Anthropic says Sonnet 5.5’s cyber capabilities improved substantially, so it launched with safeguards similar to Opus 5.5 and may fall back to Sonnet 5 for higher-risk cyber work. It also retains Sonnet’s biology safeguards and was tested in an automated behavioral audit of roughly 1,850 scenarios. Anthropic notes that no evaluation catches every failure, so routine software development and high-risk security work still need separate controls.
Sonnet 5.5 is available through the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure under `claude-sonnet-5-5`. Developers migrating from a thinking-off Sonnet configuration need to use the new `between_tools` setting. That integration detail matters as much as the model name when updating an agent runtime.
For coding-agent workflows, the practical change is lower unit cost for medium-complexity tasks, not a claim that all engineering work can be automated. Teams should re-evaluate against their own repositories, tests, permissions, and rollback process, reserving higher effort or Opus for architecture, cross-file decisions, and high-impact changes.
