Anthropic says Claude Opus 5.5 costs 40 per cent less than Opus 5

Anthropic has released Claude Opus 5.5, the first model in a new Claude 5.5 family. In its 22 September 2026 announcement the company says the model performs at the level of Claude Fable 5.1, its larger model, on most work, and costs 40 per cent less to run than Opus 5 on typical workloads at default settings.[1]

It is the company's first release since it called for pacing the frontier, and external evaluators including METR tested it before release. Claude Sonnet 5.5 and Claude Haiku 5.5 are to follow in the coming weeks, Anthropic says.[1]

The saving has two parts. The first is a lower rate card: US$4 per million input tokens and US$20 per million output tokens, against US$5 and US$25 for Opus 5. Cache reads, which Anthropic says make up most of the cost of agentic and coding work, fall from US$0.50 to US$0.20 per million tokens.[1][2]

The second is usage. Anthropic says Opus 5.5 needs fewer tokens per task, and that the two effects together net out to the 40 per cent drop. That combined figure comes from the company's own tests. Its public pricing page confirms the new rates.[1][2]

On Anthropic's benchmark table Opus 5.5 leads Opus 5 across agentic coding, computer use and knowledge work, scoring 66.4 per cent against 52.3 per cent on Terminal-Bench 4.0, for example. Several comparisons with OpenAI models use figures reported by OpenAI rather than results Anthropic ran itself.[1]

The company warns that at these capability levels benchmark margins have become a less reliable guide to real-world differences. Early-tester quotes from firms such as GitHub and Deloitte report fewer steps per job, but they are selected and published by the vendor, not independent measurements.[1]

Anthropic says Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity, so it is deployed with safeguards similar to those on Fable 5.1. Organisations that want to use it for biology research must apply to a Life Sciences Verification Program, and access for verified cybersecurity practitioners is to widen in the coming weeks.[1]

The company reports the best scores of any model it has tested on its automated behavioural audit, and says Opus 5.5 resists prompt injection better than Opus 5. It also discloses a trade-off: when safeguards intervened during some benchmark runs, older models completed the task, which it says likely lowered Opus 5.5's scores.[1]

Opus 5.5 is available on Anthropic's own platform and, the announcement says, on Amazon Web Services, Google Cloud and Microsoft Azure. Anthropic is also raising five-hour usage limits on its Pro, Max, Team and seat-based Enterprise subscription plans.[1]

Like earlier Opus models it is offered with zero data retention, and it carries watermarking measures to comply with the EU AI Act. The model can no longer be run with its thinking mode switched off, which teams upgrading existing integrations should test first.[1]