Claude Sonnet 5.5 is faster, cheaper and the first Sonnet with cyber safeguards

Anthropic released Claude Sonnet 5.5 on September 28, the second model in the Claude 5.5 family after Opus 5.5. Anthropic calls it a clear upgrade over Sonnet 5: it generates output more than 30% faster and costs up to 30% less per task, because it needs fewer tokens to finish the same work.

The key facts

• Price unchanged: $2 per million input tokens and $10 per million output tokens, the same as Sonnet 5. Cache reads cost $0.20.
• Coding jump: 70.6% on Terminal-Bench 4.0, up from 10.3% for Sonnet 5. On CursorBench 4.0 it scores 55.5%, within about two points of Opus 5.5.
• Knowledge work: Anthropic says it lands close to Opus 5.5 on GDPval-AA and clearly ahead of Sonnet 5 on long-horizon tasks.
• Availability: live on the Claude Platform as `claude-sonnet-5-5` and on AWS, Google Cloud and Microsoft Azure.
• Haiku next: Claude Haiku 5.5 is due in the coming weeks.

Anthropic positions Sonnet 5.5 as the everyday workhorse: well-scoped tasks, bug fixes, and polished documents, slides and spreadsheets. Opus 5.5 stays the pick for complex, open-ended work that needs sustained judgment.

Safeguards are part of the story

Because its cyber capabilities are comparable to Opus 5's, Sonnet 5.5 is the first Sonnet to launch with cyber safeguards and fallbacks. Higher-risk security tasks visibly fall back to Sonnet 5, while normal bug finding and fixing keeps working. It is also the first Sonnet with classifiers that block reasoning extraction, which targets so-called distillation attacks.

Why it matters

The price did not move, but the cost per finished task did. For teams that run a lot of agent loops, fewer tokens and faster output add up quickly. It also shows a pattern: capable mid-tier models now ship with the same kind of guardrails that used to be reserved for the top model.

Dany's take

The headline is not the benchmark jump, it is the cost per task. Many people compare list prices, but what you pay is tokens used times price. A model that needs far fewer tokens is effectively cheaper even at the same rate. I also like that the safeguards are described openly, including the fallback to the older model. Check your own workloads before you switch, because benchmark gains do not always carry over.

Source: Anthropic, Claude Sonnet 5.5