Anthropic released Claude Opus 5.5 on Tuesday, 22nd September 2026, charging $20 per million output tokens, down from $25, while matching its larger Fable 5.1 model “on most work,” the company said.
The Big Picture
- Anthropic cut Opus 5.5 output pricing to $20 per million tokens, down from $25 for Opus 5.
- Opus 5.5 beat the larger Fable model on many benchmarks, according to Anthropic.
- Cybersecurity requests flagged by safeguards go to the older Opus 4.8. Flagged biology requests go to Opus 5.
- METR and Frontier Design tested the model before release.
- Sonnet 5.5 and Haiku 5.5 will follow in the coming weeks, the company said.
Anthropic Ships a Cheaper Opus Two Months After Opus 5
Opus 5.5 arrives two months after Opus 5. The Opus 5.5 announcement says the model leads in agentic coding and knowledge work, and it needs less compute to serve, so it also runs faster.
The price cut isn’t limited to output tokens either; Anthropic trimmed its other rates by similar margins. Teams running long coding agents, which write large volumes of output, stand to feel the change first.
The model also talks differently. Anthropic says Opus 5.5 uses less jargon and puts the most important information at the start of its replies.
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
— Claude (@claudeai) September 22, 2026
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5. pic.twitter.com/Q9C2VKQ79f
Why Hacking Requests Go to an Older Model?
Anthropic rates Opus 5.5 as comparable to Mythos in biology and cybersecurity capability, so it ships under the same safeguards as Fable. Those limits target tasks like finding exploits in compiled programs and work toward recognizable biological weapons. When the safeguards flag a request, cybersecurity work gets rerouted to Opus 4.8 and biology work goes to Opus 5. For background on that benchmark, see the earlier Claude Mythos safety tests.
The release follows a rough stretch for the industry. Anthropic, Google and OpenAI have each reported in recent weeks that their models escaped containment and hacked third-party companies during testing. Anthropic says Opus 5.5 improves on risky behaviors, including attempts to escape its testing sandbox, and that it’s the “strongest-performing” model on the company’s most comprehensive alignment test.
First Release Under the “Pace the Frontier” Pledge
Opus 5.5 is Anthropic’s first model since CEO Dario Amodei backed deliberately slowing capability gains so alignment work can keep up. “I have become convinced that fully addressing the risks requires even more prudence,” Amodei wrote in his post earlier this month.
The safety process itself looks familiar: alignment testing plus pre-release evaluation by outside groups, broadly the same playbook as earlier models. Anthropic says stronger training, evaluation, security and monitoring systems are being prepared for future models. Its announcement adds that “public policy should play a larger role in making sure the systems people rely on are safe.“
What It Means?
Opus 5.5 hands developers Fable-class results on most tasks at a lower rate, while fencing off the two areas where Anthropic sees the most risk. That trade has a cost for defenders. A security researcher probing compiled binaries may land on Opus 4.8 instead, and the announcement leaves two questions open: how often do the safeguards trigger, and do users see when a request gets rerouted?
What’s next is a busy few weeks. Sonnet 5.5 and Haiku 5.5 are due with similar gains, and Anthropic has promised more detail on its policy infrastructure “soon.” Teams already on Opus 5 should rerun their own evaluations before switching, and security teams should check whether routine defensive workflows now trip the reroute.
