skip to content
The Weighted Average

Wire

Claude Sonnet 5.5 costs up to 30% less per task

Anthropic launched Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens, reporting up to 30% lower cost per task and 30%+ faster generation than Sonnet 5. Its vendor-reported Terminal-Bench 4.0 score is 70.6%, versus 10.3% for Sonnet 5, but the savings depend on using fewer tokens and the right effort setting. Teams with instrumented Claude workloads should put it on matched coding and document evals; the archive’s Opus 5.5 migration audit shows why model price alone cannot predict request cost.