Claude Sonnet 5.5 launched today and is now available in Ana.
The biggest gains are in agentic coding. Anthropic reports 70.6% on Terminal-Bench 4.0, up from Sonnet 5's 10.3%. Early testers also found it batches tool calls together, completing work in fewer steps.
Its cybersecurity capabilities are comparable to Opus 5, according to Anthropic. It's the first Sonnet to ship with similar cyber safeguards to Anthropic's most capable models; routine software development is unaffected.
For data work, it approaches Opus 5.5 on knowledge-work and chart-reading benchmarks. Anthropic also reports over 30% faster output and up to 30% lower cost per task than Sonnet 5. Token prices are unchanged; it uses fewer tokens to finish the work.
In our internal benchmarks, Claude Sonnet 5.5 scores 51.8 at its highest effort setting, roughly triple the 17.3 scored by Claude Sonnet 5. That jump moves Sonnet into the same performance range as the strongest models we test, at about $3.74 per task. Claude Opus 5.5 still scores higher at 59.5, but it costs more, around $4.71 per task. For most teams, Sonnet 5.5 now delivers about 87% of Opus 5.5's score at roughly 80% of the cost.
Try it by opening a new thread and selecting Sonnet 5.5 from the model picker.
Enabling Access
To enable Sonnet 5.5 for your organization, go to Settings > Models and toggle Enabled next to Claude Sonnet 5.5.
Full model configuration options are in our documentation.
