Claude Sonnet 5.5 Makes the Mid-Tier Model the Default

Claude Sonnet 5.5 scores 70.6 percent on Terminal-Bench 4.0, up from 10.3 percent for Sonnet 5 — and Anthropic says it runs more than 30 percent faster and costs up to 30 percent less per task. Per-token pricing did not move: $2 per million input tokens and $10 per million output, with cache reads at $0.20 and cache writes at $2.50. The savings come from the model finishing work in fewer tokens, not from a price cut.

The knowledge-work number matters more for most companies than the coding one. On the GDPval-AA benchmark, Sonnet 5.5 lands just two points below Opus 5.5, the flagship. When the mid-tier model sits that close to the top of the line, the default choice for production automation changes: you reserve the expensive model for the few tasks that need it and run everything else on Sonnet.

The pricing history supports that reading. Sonnet 5 launched on June 30 at introductory rates of $2 and $10 per million tokens through August 31, with a scheduled step up to $3 and $15 afterward, plus a new tokenizer that maps the same input to roughly 1.0 to 1.35 times more tokens. Sonnet 5.5 is listed at $2 and $10 with faster, cheaper task completion on top.

Safety is the other change worth reading. This is the first Sonnet model shipping with cyber safeguards comparable to Opus 5.5. It keeps Sonnet 5’s biology safeguards, adds classifiers against distillation attacks, and expands protections around preserved thinking. It is available on the Claude Platform and through AWS, Google Cloud, and Microsoft Azure as claude-sonnet-5-5.

The practical move for anyone running agents in production: re-run your own evaluation set against Sonnet 5.5 before you renew a spend commitment on the larger model. A benchmark gap of two points is a hypothesis about your workload, not a result. Test it, and ship the cheaper path where it holds.