Claude Sonnet 5.5 makes an important distinction visible: a cheaper task does not always require a cheaper token. In its September 28 release, Anthropic kept Sonnet’s published input and output rates unchanged while claiming that the model needs fewer tokens to finish the same work. That makes efficiency, rather than the price sheet alone, the useful question for teams choosing an everyday model. Anthropic’s release
The published rates are $2 per million input tokens, $10 per million output tokens and $0.20 per million cache-read tokens. Anthropic says output generation is more than 30% faster than Sonnet 5 and that its tests found task costs up to 30% lower. The phrase “up to” matters: these are vendor measurements, not a guarantee for a particular workflow. A shorter response is valuable only if it preserves the required result.
The release positions Sonnet 5.5 for well-scoped everyday tasks, bug fixes and polished documents, slides and spreadsheets. Anthropic also says Opus 5.5 remains stronger on complex, open-ended work requiring sustained judgment. My inference is that the sensible comparison is a routing decision: which tasks have clear acceptance criteria, and which need a model to resolve ambiguity over a long session? A strong benchmark score cannot make that distinction for a team.
Effort settings further complicate the cost comparison. The default is Medium in Claude apps and Claude Code, but High on the Claude Platform. Comparing bills without accounting for that setting can confuse a product default with a model efficiency gain.
Availability also has boundaries. Anthropic says Sonnet 5.5 is available across its platforms, including AWS, Google Cloud and Microsoft Azure, with zero data retention available. Higher-risk cybersecurity work can visibly fall back to Sonnet 5 under the new safeguards.
For buyers, the practical takeaway is to compare cost per accepted result at the intended effort setting. The launch provides a reason to reconsider everyday model defaults; it does not establish universal savings or remove the need to reserve more capable models for demanding work.