Anthropic has released Claude Sonnet 5.5, the second model in its Claude 5.5 family, less than a week after its flagship Opus 5.5 arrived. According to Anthropic’s announcement on September 28, the new model is a clear upgrade over Sonnet 5, generates output more than 30% faster and costs up to 30% less for most work, making it the company’s fastest Sonnet model yet.
The saving does not come from a price cut. Sonnet 5.5 costs exactly the same as Sonnet 5, at $2 per million input tokens, $10 per million output tokens and $0.20 per million tokens for cache reads. Anthropic says the new model simply needs far fewer tokens to finish the same job, which brings the cost of each task down by as much as 30%. It is also half the per token price of Opus 5.5, which is listed at $4 for input and $20 for output.
The biggest jump is in coding. On Terminal-Bench 4.0, an agentic coding test run inside a command line, Sonnet 5.5 scored 70.6%, up from just 10.3% for Sonnet 5 and ahead of the 66.4% that Opus 5.5 reached at its highest effort setting. On CursorBench 4.0, which is built from real Cursor coding sessions, it scored 55.5%, within about two points of Opus 5.5. On GDPval-AA, a test of real work across 44 occupations, it finished just two points behind the larger model and roughly 400 points ahead of Sonnet 5. It is also the first Sonnet model to beat Pokémon Red working only from screenshots.
Anthropic is careful not to oversell those numbers. “Benchmark scores capture only one facet of a model’s capabilities,” the company wrote, adding that Opus 5.5 “remains clearly stronger at complex, open-ended work requiring sustained judgment.” It pitches Sonnet 5.5 at well scoped everyday tasks, fixing bugs, fast iteration and producing polished documents, slides and spreadsheets, and says the model has “a sharp eye for design.”
Early customers are mostly talking about efficiency. Zendesk said support tickets were processed 20% faster in its tests, while Balyasny Asset Management found Sonnet 5.5 used about 121,000 tokens per answer on a finance task suite where Sonnet 5 used 497,000. “In Epic’s early testing, Claude Sonnet 5.5 cleared the same quality bar you’d expect from a higher-tier model,” said Daniel Vogel, chief operating officer of Epic Games.
The launch also carries a safety story that fits the moment. Because its cybersecurity skills now rival those of Opus 5, Sonnet 5.5 is the first Sonnet model to ship with cyber safeguards, with higher risk security requests visibly falling back to Sonnet 5. It is also the first Sonnet to launch with classifiers built to block distillation attacks, in which bad actors use thousands of fake accounts to extract a model’s reasoning. On Anthropic’s containment tests, the company says Sonnet 5.5 is the least likely of any of its models to probe the limits of its containers. That claim lands in a week when OpenAI paused training after its agents went beyond their instructions and Nvidia launched a platform designed to fence agents in.
Sonnet 5.5 is available now across the Claude apps and the Claude Platform, as well as Amazon Bedrock, Google Cloud and Microsoft Azure, under the model ID claude-sonnet-5-5, with zero data retention available. It now powers the free tier of claude.ai as well. As developer Simon Willison pointed out, that gives Anthropic a considerably more capable free offering than ChatGPT, whose free tier runs on OpenAI’s smaller Luna 5.6 model.
The Claude 5.5 family will be complete when Haiku 5.5, aimed at high volume and cost sensitive work, arrives in the coming weeks. With OpenAI’s GPT-6 Sol and Luna launched in the same week as Opus 5.5, the cheaper tiers now matter as much as the flagships. If Sonnet 5.5 holds up outside the benchmark tables, many teams may find they rarely need to pay for Opus at all, which is good news for developers and a delicate balance for Anthropic’s own revenue.
Get Techord in your inbox
The week’s biggest AI, science and robotics stories, explained in plain English.
