Anthropic says Sonnet 5.5 cuts task costs by using fewer tokens
The model is listed at $2 per million input tokens and $10 per million output tokens; Anthropic says it runs 30%+ faster and costs up to 30% less per task.
By Ryan Merket · Published
Primary source: Anthropic
Why it matters
Anthropic's claim is about lowering the cost of completed work, not cutting token prices. For agentic workloads, fewer steps and faster output could make routine automation more economical, while Sonnet's reported cyber capabilities bring Anthropic's safeguards into its mid-tier model.
Dario Amodei's Anthropic released Claude Sonnet 5.5 on September 28th, pitching the model as a faster, more economical option for everyday coding and office work. Anthropic says Sonnet 5.5 can complete most work with fewer tokens, lowering the cost of a task. TechCrunch reported the release the same day.
Dario Amodei's biography says he led research at OpenAI, where he helped direct work on GPT-2 and GPT-3, and earlier worked at Google Brain. He holds a doctorate in biophysics from Princeton and did postdoctoral research at Stanford School of Medicine. Anthropic lists Amodei as its co-founder and CEO, responsible for research direction and strategic vision.
Anthropic says Sonnet 5.5 generates output at least 30% faster than Sonnet 5 and costs up to 30% less per task for most work. Anthropic's launch page lists API prices of $2 per million input tokens and $10 per million output tokens. The company-reported savings claim rests on using fewer tokens to do the same work. For this release, Anthropic is emphasizing task-level cost alongside token prices.
The bill is about the whole task
For agentic workloads, total token use and tool-call count can matter more than the price of one token. A coding agent may inspect files, run commands, call tools, revise its work and repeat the cycle. In those settings, token price is only one part of the bill; the number of steps and how much the model writes can push the total cost up. Anthropic says early testers found that Sonnet 5.5 batched tool calls together more often than Sonnet 5, leading to fewer steps and lower costs. If that pattern holds in customers' workflows, a stable per-token price can still mean a lower bill for each completed task.
The release follows Sonnet 5's June 30th launch, which Anthropic also positioned around agentic work at lower cost. The new model's performance claim is more ambitious: Anthropic's benchmark table reports Sonnet 5.5 at 70.6% on Terminal-Bench 4.0, compared with 10.3% for Sonnet 5. It also reports 66.4% for Opus 5.5 on that evaluation. Anthropic published these results; independent confirmation of customer performance is not available. The launch page cautions that benchmark scores capture only part of a model's capabilities and says Opus remains stronger on complex, open-ended work requiring sustained judgment.
Anthropic presents Opus as the model for harder work that demands careful judgment, and Sonnet as a quicker, less costly fit for well-scoped tasks such as fixing bugs and producing documents, slides and spreadsheets. Anthropic says Sonnet 5.5's lower task cost can make it practical to run agents or sub-agents more often. Its pitch puts task-level cost alongside benchmark performance.
More capability, more safeguards
Sonnet 5.5's claimed cybersecurity capabilities come with additional safeguards. Anthropic says those capabilities are comparable to Opus 5's, a level it says warrants safeguards and fallbacks for higher-risk cyber requests. Routine software development remains available; Anthropic says some higher-risk requests will fall back to Sonnet 5. It has also added safeguards against attempts to extract the model's reasoning at scale.
Longer-running tool use can make a model more useful and increase the consequences of misuse. Anthropic says its automated behavioral audit covered roughly 1,850 scenarios and that Sonnet 5.5 improved on or matched Sonnet 5 on most measures. These are Anthropic's own evaluations; they do not establish how the safeguards will perform across every real-world deployment.
Sonnet 5.5 is available across Anthropic's platforms and through Amazon Web Services, Google Cloud and Microsoft Azure, according to the release. Anthropic also says a Haiku 5.5 model aimed at high-volume, cost-sensitive work will join the lineup in the coming weeks, without giving a date. Buyers can test whether fewer tokens and faster output reduce the cost and time required to complete their actual work.