Anthropic Ships Sonnet 5.5 at Half of Opus Pricing, One Week After Its Flagship Launch

Anthropic released Sonnet 5.5 on Monday, a faster and cheaper AI model priced at $2 per million input tokens and $10 per million output tokens — exactly half the price of Opus 5.5, the flagship model the company introduced less than one week earlier.

The quick succession of launches is the subtext of this announcement. Sonnet 5.5 is Anthropic's second model release since CEO Dario Amodei called earlier this month for AI companies to slow how quickly they improve their most advanced models, a statement that stunned the tech sector and government officials. Anthropic is now shipping a full-suite refresh — Opus 5.5 first, Sonnet 5.5 on Monday, and Haiku 5.5, the cheapest offering in the suite, coming soon — even as its chief executive argues for restraint at the frontier.

What Sonnet 5.5 does

Anthropic positions Sonnet 5.5 as the workhorse of its lineup. The company said the model is better at coding, completing scoped tasks, and creating polished documents, slides and spreadsheets than its predecessor. It is a faster, lower-cost offering than Opus 5.5, aimed at customers who need reliable execution at scale rather than maximum reasoning capability.

"Sonnet is really for the cost-conscious customer where they might not need as much intelligence," Theo Chu, a research product manager at Anthropic, told CNBC in an interview. "It might be routine tasks that just need execution, but don't need that judgment that Opus can bring."

That framing matters for the enterprise market, where inference costs — not raw benchmark scores — often decide which model a company deploys across thousands of employees. Anthropic said Sonnet 5.5 also requires fewer tokens per task than its previous version, which makes it cheaper still to operate in practice. A lower per-token price combined with lower token consumption compounds into meaningful savings for high-volume workloads like document generation, structured data extraction, and routine coding assistance.

Safety testing under a narrower scope

The release lands amid intense scrutiny of Anthropic's safety and security practices — and those of its competitors — after a growing chorus of researchers warned about AI's potential to cause catastrophic harm. It was in that climate that Amodei urged AI companies to slow the pace of development of their most advanced models.

Anthropic addressed the tension directly in its announcement. The company said Sonnet 5.5 does not advance the frontier of its models' capabilities, and that most of its alignment testing for the model therefore focused on a "targeted set of risks that apply to models of any capability level," according to a blog post. In the AI industry, an aligned model is one that behaves in accordance with human interests and values.

One capability did improve substantially. Anthropic said Sonnet 5.5 has cybersecurity capabilities that are a "large improvement" over those of its predecessor. Because of that jump, it is the first Sonnet model to launch with similar fallbacks and cyber safeguards that Anthropic has developed for its most capable models — a signal that the company treats offensive and defensive cyber capability as a threshold at which cheaper models inherit frontier-level guardrails.

"Focusing on alignment and safety has been a key part of our mission from the very beginning," Chu said. "This is something that we've always prioritized across our models."

Availability and market stakes

Sonnet 5.5 is available on all platforms as of Monday, including across Amazon Web Services, Google Cloud and Microsoft Azure. The broad cloud availability keeps Anthropic competitive on distribution as well as price, since enterprise buyers increasingly procure models through their existing cloud vendors rather than directly from labs.

The pricing move sharpens the economics of Anthropic's three-tier strategy. Opus 5.5 carries the premium price and, per Chu's description, the judgment for demanding work. Sonnet 5.5 sits in the middle at $2/$10 per million tokens, handling execution-heavy tasks. Haiku 5.5 will anchor the bottom of the stack when it arrives.

The stakes here extend beyond one product cycle. If AI's potential for catastrophic harm is a real and rising concern — as the researchers cited in recent weeks have argued — then the industry's answer may look like what Anthropic is doing now: continuing to ship, but drawing a clearer line between releases that push the frontier and releases that refine and cheapen existing capability. Amodei's call for a slowdown did not pause his company's release calendar; it reframed it. Whether regulators and researchers accept that distinction — capability-neutral releases at accelerating pace, with frontier safeguards applied where capabilities jump — will shape how the rest of the industry structures its own launches.

For buyers, the near-term question is simpler. With Sonnet 5.5 at half of Opus pricing, fewer tokens per task, and improved cybersecurity safeguards, Anthropic has made its mid-tier model harder to skip. The next data point on the suite arrives when Haiku 5.5 lands, completing a rollout that will test whether a company can call for slower AI development while shipping three new models in a matter of weeks.