Anthropic launches Claude Sonnet 5.5 with faster, lower-cost model option

The new model retains Sonnet 5’s standard input and output token prices. Anthropic attributes its lower per-task costs to reduced token consumption and cheaper cache reads.

Anthropic released Claude Sonnet 5.5 on 28 September 2026, positioning it as a faster option for coding, everyday knowledge work and repeatable AI tasks. The company says the model generates output more than 30% faster than Claude Sonnet 5 and costs up to 30% less per task in its testing, depending on the workload.

In its official model announcement, Anthropic describes Sonnet 5.5 as a lower-cost complement to Claude Opus 5.5, which remains aimed at more complex assignments requiring sustained judgment. The savings claim concerns the work completed, rather than a blanket reduction in the advertised price of input and output tokens.

Lower task costs without lower base token rates

Sonnet 5.5 keeps Sonnet 5’s standard API pricing: $2 per million input tokens and $10 per million output tokens. Cache reads fall to $0.10 per million tokens from $0.20 for Sonnet 5. Anthropic attributes additional savings to the new model using fewer tokens to complete comparable tasks, rather than a lower output-token rate.

Applications with different prompt lengths, caching patterns or amounts of model reasoning may see different results. Anthropic also offers effort controls: Medium is the default in Claude applications and Claude Code, while the Claude Platform defaults to High. Raising effort can improve results on harder tasks but changes token consumption and cost.

Coding and office work are the intended middle ground

Anthropic identifies bug fixing, routine code changes, document creation and slide production as suitable work for Sonnet 5.5. Its published evaluation gives the model 70.6% on Terminal-Bench 4.0, compared with 10.3% for Sonnet 5. Those are results reported by Anthropic for a particular agentic coding evaluation, not a guarantee of comparable performance in production.

The company also quotes early customers testing the model. Slack, for example, reported approximately 14% fewer output tokens than Sonnet 5 across its offline Slackbot evaluations. These results come from customer-specific tests presented in Anthropic’s launch material; they do not establish a general saving for all business applications.

Available in Claude and developer platforms

Anthropic says Sonnet 5.5 is available in Claude on the web and mobile, through its API under the model identifier claude-sonnet-5-5, and through AWS, Google Cloud and Microsoft Foundry. Developers therefore have deployment routes beyond the consumer chat application.

Moving existing applications is not necessarily a model-name change alone. Anthropic’s migration guide says Sonnet 5.5 rejects forced tool_choice settings that worked with earlier models; developers should use automatic tool selection and strict tool schemas where appropriate. Applications that previously disabled thinking must also change their configuration to the supported between_tools setting. These compatibility requirements matter for teams considering the company’s stated speed and cost improvements in existing workflows.