Anthropic · Model release
Claude Sonnet 5.5: everyday coding with breaking API changes
Sonnet 5.5 is released at Sonnet 5's $2/$10 token price, with 1M context and 128K standard output.
What changed
- In the CursorBench 4.0 snapshot reviewed 2026-10-01: 55.5% at Max effort and $9.67/task; 47.8% at High and $1.67/task.
- Terminal-Bench 4.0: Anthropic reports 70.6%; Artificial Analysis's independently run mini-swe-agent result reviewed 2026-10-01 is 63.6% at max. These are different evaluation setups.
Best fit
- Daily engineering
- Document and slide work
- Anthropic-stack integrations
What to be careful about
- Adaptive thinking is on by default. Use between_tools to turn off up-front thinking; forced tool use returns an error.
- 300K output is a Message Batches beta, not the standard output limit. Haiku 5.5 remains announced, not released.
Dataset snapshot: Claude Sonnet 5.5 is also represented in the checked comparison dataset. Prices and benchmark figures carry their own field dates.