Summary

Anthropic says Sonnet 5.5 is over 30% faster and costs up to 30% less per task than Sonnet 5. It scored 70.6% on Terminal-Bench 4.0, compared with 10.3% for Sonnet 5, while retaining a 1-million-token context window.

Why it matters

A stronger mid-tier model at unchanged token pricing could shift routine coding-agent workloads away from more expensive frontier models.

Sources