Anthropic launches Sonnet 5.5 with a 30% speed claim

Sonnet 5.5 targets faster everyday AI work

Anthropic has released Sonnet 5.5, the latest version of its mid-tier model, and says it is 30% faster than Sonnet 5. The company is positioning the model as an assistant for routine work, including coding and the creation of office documents.

Speed is the central claim of the release. Anthropic also says Sonnet 5.5 burns through tokens at a significantly slower rate than its predecessor, presenting the combination as a way to make the model a cheaper and faster work partner.

Sonnet 5 was announced about three months earlier with an emphasis on efficient agentic deployment. In that context, Sonnet 5 agent efficiency and pricing outlines the earlier focus on lower-cost agent tasks and a narrower gap with top-tier models.

Agentic coding and model positioning

Within Anthropic’s range, Sonnet sits below Opus in overall capability. The company nevertheless argues that Sonnet can be more useful in particular situations because of its agility.

Anthropic’s benchmarks indicate that Sonnet 5.5 performs better than Opus 5.5 on agentic coding. The company attributes that result to Sonnet 5.5’s ability to spawn multiple agents without exceeding cost limits. The claim is material for organisations that use agentic systems for software work, where model selection involves both capability and operating cost.

Cyber safeguards extend to the new Sonnet

Anthropic also says Sonnet 5.5 has cyber capabilities comparable to Opus 5. It is therefore the first Sonnet model to be covered by the same cyber safeguards that apply to Fable and Opus.

The company said it plans to introduce a new version of Haiku, its smallest model, in the coming weeks, but did not give a release date. The announcement arrives amid a rapid stream of releases from major AI labs, including new OpenAI models and a Meta model intended to power an upcoming smart-glasses feature.

For businesses, the practical next step is to assess Sonnet 5.5 against existing coding and document workflows, measuring its claimed speed and token efficiency while applying the cyber controls appropriate to its stated capabilities.