Anthropic launches Sonnet 5.5 with 30% faster responses and lower running costs

Anthropic released Sonnet 5.5 on September 28, a mid-tier version of its Claude model line that the company says runs 30% faster than its predecessor while costing significantly less to run. Anthropic pitches the model as an everyday work partner for tasks such as coding and drafting office documents.

Sonnet 5.5 replaces Sonnet 5, which the company introduced about three months earlier. That version was sold mainly on its ability to run agents — AI systems that carry out multi-step tasks on their own — at a lower cost than rival models. With 5.5, Anthropic says it has kept that cost advantage and added speed, along with a slower rate of token burn, the pace at which a model consumes the text chunks it processes to generate a response.

In Anthropic’s internal hierarchy, Sonnet sits below Opus, the company’s most capable model, but the smaller model’s speed can make it more practical for certain jobs. Anthropic’s own benchmarks show Sonnet 5.5 outperforming Opus 5.5 on agentic coding tasks, which the company attributes to Sonnet’s ability to run multiple agents at once without breaching cost limits.

Anthropic also says Sonnet 5.5 now has cyber capabilities “comparable” to Opus 5, the prior flagship model. That claim carries a practical consequence: Sonnet 5.5 becomes the first Sonnet model to fall under the same cyber safeguards Anthropic applies to its Opus and Fable models, restrictions meant to limit misuse of a model’s hacking-related abilities.

The company said a new version of Haiku, its smallest and cheapest model, is coming within a few weeks, though it has not set a release date.

The release lands amid a wave of updates from rival labs. Last week OpenAI put out upgraded versions of Sol and Luna, its mid-tier and budget models, and Meta announced a new model to power an upcoming feature tied to its smart glasses. Whether Sonnet 5.5’s speed and cost claims hold up will depend on independent testing once developers get broader access, and on how the added cyber safeguards affect its behavior in practice. Anthropic’s promised Haiku update, still without a firm date, is the next release to watch.