The lede
Introducing Claude Sonnet 5.5
Anthropic
The headline number is the one that matters for anyone running a fleet: 70.6% on Terminal-Bench 4.0 against Sonnet 5's 10.3%, two points off Opus 5.5 on GDPval-AA, at the same $2/$10 per million tokens but typically needing far fewer of them — Anthropic's own testing puts it up to 30% cheaper per task and 30%+ faster. The quieter news is that it ships with the cyber safeguards previously reserved for the top of the family, because its security capabilities now match Opus 5's, and that it is the model behind the free tier on claude.ai, which is a more interesting competitive move than any of the bars on the chart.