Anthropic launches Opus 5.5, cuts prices and tightens safety guardrails

Anthropic released Opus 5.5 on Tuesday, the newest version of its most capable and most expensive Claude model. The company says the model sets a new standard for coding and knowledge work, and that it beat Anthropic’s own larger Fable model on many benchmarks and in a number of informal tests Fable failed to complete.

Opus 5.5 also costs less to run. Output tokens, the text a model generates in response to a prompt, are priced at $20 per million, down from $25 for the previous Opus. Other pricing tiers dropped by similar margins, and Anthropic says the model runs faster, a result of lower compute needs to serve it.

The company also changed how the model communicates. Opus 5.5 uses less jargon than earlier versions and puts key information at the start of its responses rather than at the end.

The release comes about two months after Opus 5, which launched July 24. Anthropic says Sonnet 5.5 and Haiku 5.5, the two lower tiers in its three-model lineup, will follow “in the coming weeks” with similar gains in performance and price.

Same guardrails as the larger model

Anthropic says Opus 5.5 matches its Mythos model in biology and cybersecurity risk, so it ships under the same safety limits placed on Fable. Those limits restrict how much the model can be used to find exploits in compiled software or to help design biological weapons.

The release is the first since chief executive Dario Amodei said Anthropic would deliberately slow the pace of capability gains to let safety work keep up. “I have become convinced that fully addressing the risks requires even more prudence,” Amodei wrote in a post earlier this month, “not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up.”

Anthropic says the safety testing behind Opus 5.5 was broadly similar to earlier releases, including outside evaluation from groups such as METR and Frontier Design. The company said it is already building more advanced training and monitoring systems for future models, without giving details.

“As AI becomes more capable, public policy should play a larger role in making sure the systems people rely on are safe,” Anthropic wrote in its announcement. “That capacity takes time to build, and we’ve started to put the infrastructure in place to support it. We expect to share more details on these efforts soon.”

What that infrastructure looks like, and how it applies to Sonnet 5.5 and Haiku 5.5, is not yet public. Anthropic has not set a date for either release beyond “coming weeks,” and has not said what specific policy or monitoring changes it plans to announce next.