Anthropic released Claude Sonnet 5.5 on September 28 across its apps and developer platform, alongside Amazon Web Services, Google Cloud and Microsoft Azure. The company positions it for everyday coding and document work, while retaining Opus 5.5 for more complex, open-ended tasks requiring sustained judgment.
Advertisement
Base API prices remain $2 per million input tokens and $10 per million output tokens, with cache reads at $0.20 per million tokens. Anthropic’s claim of lower task costs comes from using fewer tokens in its evaluations, rather than a reduction in those base rates. The company also reports output generation more than 30% faster than Sonnet 5. Neither claim guarantees the same saving on a particular production workload.
Effort settings and migration
Anthropic says reasoning effort changes the balance among speed, cost and task quality. The default is Medium in Claude Code and its apps, and High on the Claude Platform. Comparing model costs without accounting for those settings can obscure the actual work each model performs.
Developers can call the model as claude-sonnet-5-5. Workloads that previously used Sonnet with thinking disabled need to move to the new between_tools setting, which keeps up-front thinking off. The launch page links a migration guide for that change. Anthropic also says Sonnet 5.5 remains available with zero data retention, as were Sonnet 5 and Opus 5.5.
On Terminal-Bench 4.0, Anthropic reports 70.6% for Sonnet 5.5, compared with 10.3% for Sonnet 5 and 66.4% for Opus 5.5. These are benchmark-specific results. A separate footnote concerns GDPval-AA and AA-Briefcase: those evaluations used a pre-release deployment affected by a structured-output bug that Anthropic says it has since fixed. That caveat does not describe the Terminal-Bench result.
How safeguards affect use
Sonnet 5.5 is the first Sonnet release with cyber safeguards and fallbacks of the kind Anthropic uses on its more capable models. The company says higher-risk cyber tasks visibly fall back to Sonnet 5, while routine development remains available. It also extends preserved thinking, which binds reasoning traces to the account that created them and can matter when moving conversations between accounts.
The model is the second release in the Claude 5.5 family. Anthropic announced plans for a Haiku 5.5 model aimed at high-volume, cost-sensitive work in the following weeks, without giving a launch date.