Anthropic has introduced Claude Sonnet 5.5, a new version of its Sonnet model designed for practical everyday tasks, software development and AI agent workflows.
According to Anthropic, Sonnet 5.5 is 30% faster than the previous Sonnet 5 model and can reduce costs by up to 30% across many tasks. The model is designed to handle tasks such as fixing code, working with documents and spreadsheets, and completing multi-step workflows through AI agents.
The biggest improvement can be seen in software development and agentic tasks. In the Terminal-Bench 4.0 test, Sonnet 5.5 achieved a score of 70.6%, compared with 10.3% for Sonnet 5. Anthropic says the new model can also call multiple tools at the same time, reducing the number of intermediate steps required for complex tasks.
Anthropic has kept the model’s API pricing unchanged at $2 per million input tokens and $10 per million output tokens, while cache reads cost $0.20 per million tokens. However, the company says Sonnet 5.5 uses fewer tokens to complete the same tasks, which can lower the overall cost of individual workloads.
The company has also upgraded cybersecurity safeguards to the level used for its Opus models and introduced additional restrictions for potentially harmful requests.
Claude Sonnet 5.5 is now available through Claude applications and Anthropic’s developer platform. The smaller Claude Haiku 5.5 is expected to launch in the coming weeks.
