On 28 September Anthropic released Sonnet 5.5, the second model in the 5.5 family. Per the company it is over 30 percent faster than Sonnet 5 and up to 30 percent cheaper per task, at the same 2 and 10 dollars per million tokens. It is also the first Sonnet to launch with cyber safeguards.
- The sticker price has not moved. The saving comes from fewer tokens per task.
- On the terminal benchmark Terminal-Bench 4.0 it jumps from 10.3 to 70.6 percent, per Anthropic.
- Its cyber capabilities are comparable to Opus 5, so higher-risk cyber tasks visibly fall back to Sonnet 5.
The sticker price has not moved. Two dollars per million input tokens, ten per million output. What has changed is how many tokens it needs to do the same thing.
That difference rarely makes the headlines, and it is exactly what reaches the bill. A model at the same price that gets to the answer with less wandering is a cheaper model. It just does not say so on the price list.
The number that catches the eye is the terminal one: from 10 to 70 percent. A jump like that does not come from polishing; it looks like a model that has learned a new trade.
Otherwise Anthropic says itself where the ceiling is. On complex, open-ended work that needs sustained judgment, Opus 5.5 remains clearly stronger. Sonnet is for well-scoped everyday work: a bug, a document, a spreadsheet, slides.
And one detail more interesting than the benchmarks. Until now Sonnet was the middle model that did not need cyber brakes. Now it does. When a company's middle model reaches the cyber capability of Opus 5, released in July, the brake moves down a floor.
If you pay for Sonnet 5 in production, one caution from the announcement: if you run it with thinking off, you need to move to the new between_tools setting before switching.