Claude Sonnet 5.5 Model Description and Safety Evaluations Released

Anthropic News has released the description and safety evaluations of Claude Sonnet 5.5, a language model designed for everyday tasks, bug fixing, and document creation. According to the release, Claude Sonnet 5.5 is a faster, lower-cost complement to Claude Opus 5.5 and is strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets. The model can understand both text and image inputs, engaging in conversation, analysis, coding, and creative tasks, and can output text, including text-based artifacts, and diagrams. Claude Sonnet 5.5 was pretrained on large, diverse datasets to acquire language capabilities and underwent substantial post-training to make it an effective assistant whose behavior aligns with the values described in Claude's constitution.

The model was trained on a proprietary mix of publicly available information from the internet, public and private datasets, and other sources, including synthetic data generated by other models. Claude Sonnet 5.5 has been deployed with the same chemical and biological misuse protections as Claude Opus 5, treating it as having CB-1 capabilities. The model has undergone safety evaluations, including assessments of its even-handedness in responding to requests on 150 political topics and its ability to acknowledge other points of view. The evaluations showed that Claude Sonnet 5.5 was substantially more even-handed than Claude Sonnet 5, scoring 97.9% on the API and 99.0% on claude.ai.

The model also underwent testing for honesty, including evaluations of its ability to avoid misleading users and making up facts. Claude Sonnet 5.5 did better than Claude Sonnet 5 on most of these tests, with some exceptions noted below.