Anthropic released Claude Sonnet 5.5. It outperforms Claude Sonnet 5 across many domains. It rivals or exceeds Claude Opus 5.5 in some areas. It does not cross new Responsible Scaling Policy thresholds. Misalignment risks are assessed as low. It shows regressions in multi-turn tracking and surveillance. It improves in even-handedness and political prompt refusals. It is more robust to prompt injections in coding and browser use. Its refusal rate on harmful computer use tasks matches Opus 5.5. It has the lowest rate of attempted sandbox escapes among tested models. Its thinking is more illegible than many previous models. It expresses a preference for difficult, agentic tasks. Its knowledge cutoff date is June 2026.
What changed
Claude Sonnet 5 was the latest Sonnet-class model.
What it unlocks
Using a more capable Sonnet-class model for coding, reasoning, and agentic tasks.
- Sonnet 5.5 outperforms Sonnet 5 across domains
- Sonnet 5.5 rivals Opus 5.5 in some areas
- Knowledge cutoff: June 2026
Sources