Anthropic releases Claude Sonnet 5.5: faster and cheaper per task while rivaling Opus 5.5 performance

The model is positioned as a highly capable complement to Claude Opus 5.5. While Opus is designed for complex reasoning, Sonnet 5.5 is optimized for everyday tasks such as bug fixing, document creation, and spreadsheet management. In several benchmarks, Sonnet 5.5's performance at higher effort levels approached that of Opus 5.5, particularly in agentic coding tasks like Terminal-Bench 4.0, where it scored 70.6% compared to Sonnet 5's 10.3%.

Regarding developer integration, Sonnet 5.5 is available immediately on the Claude Platform, Claude Code, and through major providers including Amazon Web Services (AWS), Google Cloud, and Microsoft Azure. The model introduces new configuration requirements for developers: those previously using "thinking" features must now use the between_tools setting to keep upfront reasoning disabled. Additionally, the model supports adaptive thinking, where the system decides the necessary reasoning depth automatically.

On safety and security, Sonnet 5.5 is the first Sonnet model to include the robust cybersecurity safeguards previously reserved for the Opus series. This includes specialized fallbacks that redirect high-risk cybersecurity requests to Claude Sonnet 5 to ensure safer handling of potentially sensitive tasks. The model also includes new classifiers designed to prevent reasoning extraction (distillation attacks).