Anthropic ships Claude Sonnet 5.5, 30%+ faster at same price

Anthropic is positioning Sonnet 5.5 as the fast, low-cost counterpart to Opus 5.5, which the company published on September 22 as the first model in the Claude 5.5 family. According to Anthropic, Opus 5.5 suits complex work that needs careful judgment, while Sonnet 5.5 targets well-scoped everyday tasks, bug fixes, polished documents, slides and spreadsheets, and it is said to have a good sense of design. A Claude Haiku 5.5 aimed at high-volume, cost-sensitive uses is also planned for release within the next few weeks.

The release keeps Sonnet's API price unchanged, but Anthropic says the model needs far fewer tokens for the same work, which is why its tests found cost per task up to 30% below Sonnet 5. It says Sonnet 5.5 is the fastest Sonnet yet. In coding, Terminal-Bench 4.0 rose from 10.3% for Sonnet 5 to 70.6%, above Opus 5.5, and on GDPval-AA the gap to Opus 5.5 narrowed to 2 points, about 400 points above Sonnet 5.

Anthropic also notes the limits of those results, saying benchmarks capture only one side of a model and that independent and internal testing still finds Opus 5.5 clearly better on complex, open-ended work. This matters for the trade-off buyers face: a cheaper fast model may fit well-defined tasks, but the choice may still hinge on how much judgment a given job needs. The same release carries first-time Sonnet safeguards close to its top-tier models, which Anthropic says can mean more refusals than Sonnet 5 on harmless cybersecurity work and accidental flags in some microbiology or virology requests. The reported dip in multi-turn safety scores for tracking and surveillance, violent extremism, and hate and discrimination is also worth watching as the model reaches general users.