AI News · Models ·
Anthropic releases faster, lower-cost Sonnet 5.5
Anthropic released Sonnet 5.5, a faster, lower-cost AI model. The company said it improves coding and completion of scoped tasks, marking its second model launch since CEO Dario Amodei called for slower development of the most advanced models.
Key points
- Anthropic released Sonnet 5.5, claiming improvements in coding, scoped tasks and business document creation.
- Pricing is $2 per million input tokens and $10 per million output tokens, half Opus 5.5’s price.
- The model is available through Amazon Web Services, Google Cloud and Microsoft Azure.
- Anthropic said stronger cybersecurity capabilities warranted safeguards similar to those used for its most capable models.
- Specific speed measurements and comparative results on individual business workflows were not reported.
What happened: Anthropic released Sonnet 5.5 on Monday, introducing a faster, lower-cost AI model aimed at customers handling routine work, CNBC reported. The company said it improves on its predecessor in coding, completing scoped tasks and creating polished documents, slides and spreadsheets. The release is Anthropic’s second model launch since CEO Dario Amodei urged AI companies to slow development of their most advanced models. Anthropic said Sonnet 5.5 does not push the frontier of its models’ capabilities, positioning it instead around better execution and lower costs.
The details: Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens, half the price of Opus 5.5, according to CNBC. Anthropic also said the model requires fewer tokens per task than its previous version, reducing the cost of running it. The size of that reduction and specific speed measurements were not reported. Sonnet 5.5 became available across all platforms on Monday, including Amazon Web Services, Google Cloud and Microsoft Azure. That gives business customers access through several major cloud platforms.
Who it affects: Anthropic is targeting cost-conscious customers whose tasks need execution rather than the judgment offered by its more expensive Opus model. Theo Chu, a research product manager at Anthropic, described routine tasks as a central use case in an interview with CNBC. For business teams, the relevant distinction is between clearly bounded assignments, such as a defined coding task, and work requiring more judgment. The launch comes less than a week after Opus 5.5. Anthropic also said Haiku 5.5, the cheapest model in the suite, is coming soon, but a release date was not reported.
Background: The release arrives amid scrutiny of AI companies’ safety and security practices, following warnings from researchers about AI’s potential to cause catastrophic harm. Anthropic said most of Sonnet 5.5’s alignment testing focused on targeted risks applicable to models at any capability level. Alignment refers to whether a model behaves in accordance with human interests and values. Despite saying the model does not advance its overall capability frontier, Anthropic described its cybersecurity capabilities as a large improvement over its predecessor. It is the first Sonnet model to launch with similar fallbacks and cyber safeguards developed for the company’s most capable models.
What to watch: Teams considering Sonnet 5.5 should test its claimed cost and speed advantages on bounded coding and operational tasks while checking output quality. The reported pricing establishes a clear difference from Opus 5.5, but comparative results on individual business workflows were not reported. Whether Sonnet’s lower price and reduced token use translate into savings without sacrificing quality remains a question for those evaluations.
Our take
Teams should test Sonnet 5.5 on bounded coding and operational tasks to see whether its cost and speed advantages hold without sacrificing quality.