Anthropic ships Claude Sonnet 5.5 at Sonnet 5's price, claiming faster, cheaper runs

Anthropic released Claude Sonnet 5.5 at Sonnet 5's $2/$10 per million token price, claiming 30%+ faster output and up to 30% lower cost per task.

Anthropic released Claude Sonnet 5.5 on September 28, per TechCrunch. Pricing is unchanged from Sonnet 5 at $2 per million input tokens and $10 per million output tokens. Anthropic says the model generates output 30%+ faster and costs up to 30% less per task because it needs fewer tokens. It is pitched for well-scoped coding, bug fixes, documents and slides. For many builders, it is likely to become the everyday model.

The headline benchmark is Anthropic's own: 70.6% on Terminal-Bench 4.0, up from 10.3% for Sonnet 5. Treat it with caution. paddo.dev notes that cyber safeguards sent 1.5% of Sonnet's trials to a fallback model, against 10% for Opus 5.5, which may skew the comparison. On that author's own 20-run monorepo test at medium effort, Sonnet 5.5 broke no passing tests and cost $11.76, against $45.81 for Opus 5.5.

The effort setting changes the economics. TokenCost reports that Artificial Analysis measured about 193,000 output tokens per task at max effort, making Sonnet 5.5 pricier per task there than Opus 5.5. It also reports that a Simon Willison max-effort test used all 128k output tokens and returned nothing. It is unclear whether that is a bug. For now, stick to high or xhigh.

Default effort differs by surface. It is medium in Claude Code and the apps, and high on the API. Anthropic's migration guide flags breaking changes. For example, thinking: {"type": "disabled"} now returns a 400 error, and you use between_tools instead. The New Stack notes that on the API, fallback to Sonnet 5 for flagged cyber requests is opt-in. Without it enabled, those requests simply stop. Haiku 5.5 has no price or date yet.

Sources