Anthropic has introduced Claude Sonnet 5.5, a new model that positions the Sonnet family as a more efficient and secure option for enterprise and developer users. The release combines improved coding performance, faster response times, and lower per-task costs with safeguards previously limited to the company's frontier systems. It also arrives during a period of heightened regulatory attention under the European Union's AI Act and growing scrutiny of frontier model security.
Performance and Pricing Gains
Claude Sonnet 5.5 records a 70.6% score on the Terminal-Bench 4.0 agentic coding evaluation, compared with 10.3% for Sonnet 5 and 66.4% for the more expensive Opus 5.5 at its highest effort setting. Anthropic states the model runs more than 30% faster and costs up to 30% less per task than its predecessor. The company also reports that Opus 5.5 costs twice as much per token, making Sonnet 5.5 a competitive option for cost-sensitive deployments.
Benchmark Positioning and Limits
On GDPval-AA, a benchmark spanning 44 occupations, Sonnet 5.5 scores 1,844, narrowly below Opus 5.5 at 1,846 and well above Sonnet 5 at 1,449. Anthropic says Opus remains clearly stronger at open-ended work requiring sustained judgment. Sonnet 5.5 therefore delivers near-flagship results in many areas without matching the top model across every capability.
Safeguards and Security Controls
Anthropic has moved a set of frontier-style cybersecurity safeguards into the Sonnet tier for the first time. The company says the model's cybersecurity capabilities are comparable to those of Opus 5, so it launches with restrictions previously reserved for its most capable systems. Higher-risk cyber requests will visibly fall back to Sonnet 5, while biology safeguards remain unchanged.
Reasoning Protection and Regulatory Context
Claude Sonnet 5.5 is also the first Sonnet model to ship with classifiers that block reasoning extraction. Its preserved thinking feature ties a model's reasoning to the account that produced it, adding a layer of protection against misuse. These protections align with Article 55 of the EU AI Act, which obliges providers of general-purpose models with systemic risk to protect their systems and report serious incidents without undue delay.
Pricing Questions and Company Positioning
Anthropic describes pricing of US$2 and US$10 as the same as Sonnet 5. Earlier reporting indicated those figures were introductory rates until 31 August, after which Sonnet 5 would cost US$3 and US$15. It remains unclear whether the increase never took effect, whether the company is comparing against the introductory price, or whether the pricing structure was adjusted after that date.
Developer and Enterprise Implications
For engineering teams, the combination of a 70.6% agentic coding score and lower per-task costs may make Sonnet 5.5 suitable for high-volume automation. The fallback mechanism for higher-risk cyber requests adds a compliance-friendly control layer that many enterprise buyers will welcome. Anthropic's decision to keep biology safeguards unchanged also signals a focused approach to specific risk categories.
The launch of Claude Sonnet 5.5 signals a shift in how Anthropic balances capability, cost, and security across its model range. By bringing frontier-style safeguards and reasoning protections into a lower-cost tier, the company is responding to both enterprise demand and regulatory pressure. The main open question is pricing clarity, which could shape how customers calculate the model's claimed savings over the long term.