Anthropic Cuts Opus-Class Pricing With Claude Opus 5.5 at $4/$20 Per Million Tokens

Claude Opus 5.5 went live on 22 September 2026 priced at $4 per million input tokens and $20 per million output tokens, with cache reads at $0.20 and cache writes at $5. Anthropic also listed a fast mode at $8/$40. That is the number that matters for anyone running recurring workloads through Opus-class models, since output tokens, the ones a model generates rather than receives, typically drive the bulk of a bill for long-form coding or analysis tasks.

Opus has been Anthropic's most capable and most expensive tier, reserved for tasks where teams tolerated higher cost for better reasoning. A price cut on that tier changes the calculus for anyone who had been routing simpler jobs to a cheaper Anthropic model just to avoid Opus pricing. Engineering managers who track per-seat or per-task API spend now have a lower ceiling to model against, and that affects build-versus-buy decisions for internal tools that lean on long context windows or high output volume.

Anthropic says Claude Opus 5.5 is reachable through Claude Pro, Max, Team, and Enterprise plans, through the Claude Platform directly, and through AWS, Google Cloud, and Microsoft Foundry. That is a wide distribution footprint, covering both Anthropic's own subscription products and the three largest enterprise cloud marketplaces. The announcement does not say whether pricing is identical across all three cloud channels or whether existing Opus workloads migrate automatically to the new model id, claude-opus-5-5, versus requiring a manual switch.

For workers using Claude through a Pro or Max seat, the price change does not alter a flat subscription fee, but it does matter for contractors and freelancers billing clients based on API consumption. Policy watchers tracking compute costs as a proxy for AI adoption should note that a headline price cut on a flagship tier often signals margin confidence or competitive pressure, not charity. The real test is whether output-heavy tasks, the kind that drive most Opus spend, see meaningful savings once fast-mode surcharges and cache-write costs are factored in.

Anyone budgeting around this should pull current invoices for Opus-class usage and compare output-token volume against the new $20 rate before assuming savings. A team heavy on cache writes, now priced at $5 per million, may see less benefit than one relying mostly on cache reads at $0.20. Check whether a given AWS, Google Cloud, or Microsoft Foundry account already exposes claude-opus-5-5 as a selectable model, since availability rolling out across three separate cloud marketplaces rarely happens on the same day for every customer.

The next move belongs to procurement teams and engineering leads who track monthly Anthropic spend. They should request a rate comparison between Claude Opus 5.5 and whatever Opus-tier model preceded it on their account, broken out by input, output, and cache-write tokens. Until that comparison exists, claims of savings remain untested on any specific workload.