Claude Sonnet 5.5: Cheaper, Faster, Almost Catches Up To The Top Model

Anthropic has released Claude Sonnet 5.5, the middle model in its lineup. On the Terminal-Bench 4.0 coding test, it scored 70.6%, compared with 10.3% for the previous Sonnet 5. Text output got more than 30% faster, and the cost per task dropped by up to 30% thanks to more efficient token use. Per-token pricing stayed the same: $2 per million input tokens and $10 per million output tokens, just like Sonnet 5.

I work in Opus 5 every day, and I hand off the hardest tasks to Fable 5.1. So I know the gap between models in the same Anthropic lineup firsthand. When a mid-tier model gets this close to the top model in knowledge, and costs up to 30% less per task at the same time, that changes the math for agent runs. Before, people paid for the top model just for the quality. For repetitive coding tasks, a jump on the benchmark from 10.3% to 70.6% looks like a real improvement. It is not a cosmetic update.