Anthropic debuts Claude Sonnet 5.5, 30% faster

Anthropic is filling out its Claude family from the top down. The Sept. 22 launch of Opus 5.5, the flagship, was followed by Sonnet 5.5, the mid-tier workhorse for document generation, summarization, spreadsheets and fixing coding bugs. Because Sonnet 5.5 is built on the same foundation as Opus 5.5, Anthropic says it is the first Sonnet with notable cybersecurity capabilities; it ships with cyber safeguards and fallbacks like those developed for more capable models such as Mythos-class models and Fable. Those may trigger on prompts touching cybersecurity, biology or other sensitive content, though Anthropic says most software development and life sciences work will be unaffected. Anthropic also plans Haiku 5.5, its smallest and fastest model for high-volume, cost-sensitive applications, in the coming weeks.

Two contrasting signals sit inside this release. On performance, Anthropic says Sonnet 5.5 scored 70.6% on Terminal-Bench 4.0, an agentic coding evaluation, against the previous model's 10.3%, and landed two points below Opus 5.5 on GDPval-AA, a test of real-world occupational work. On trust, it is the first Sonnet to carry invisible text watermarking, a step Anthropic ties to the European Union's AI Act rather than to a user request.

How this lands may depend less on benchmark gains than on how buyers weigh per-task economics against governance. The same list price as Sonnet 5 combined with fewer tokens per task is what Anthropic says produces the 30% cheaper figure, so the claim is a usage-based one. Whether compliance and content teams treat the watermarking as a plus or a friction point looks like the variable that decides how quickly Sonnet 5.5 is adopted for everyday work. That tension, not raw speed, may shape the model's reception.