Anthropic Launches Claude Opus 5.5 with 40% Lower Running Costs and Million-Token Context Window

Anthropic says Claude Opus 5.5 costs 40% less to run than Claude Opus 5 for typical workloads. That’s the headline claim from the San Francisco AI company’s launch materials, and for developers and businesses already paying token-rate bills on Anthropic’s platform, it’s the number that matters most.

The model launched on 22 September 2026 and is available immediately through the Claude API, the Claude Platform, Amazon Web Services, and Google Cloud. Microsoft Foundry availability was reported in secondary coverage, though that should be treated as requiring confirmation against current Microsoft documentation. Access through the API uses the model identifier `claude-opus-5-5`.

What Anthropic Is Claiming

The company describes Claude Opus 5.5 as the first model in its new Claude 5.5 family. According to Anthropic’s release materials, the model performs at the level of Claude Fable 5.1 on most work — a notable positioning, given that Fable 5.1 sits higher in Anthropic’s current model hierarchy. The company has not pointed to independently verified benchmark results to support that claim, so it should be read as Anthropic’s own assessment for now.

Anthropic’s pricing, as listed in its Platform documentation, puts input at 4 USD per million tokens and output at 20 USD per million tokens — around £3.10 and £15.60 respectively at current exchange rates, though UK billing amounts will vary depending on the platform, applicable taxes, and the rate on the day of billing. That represents a 20% reduction per token compared with Claude Opus 5, and the company says the combination of lower rates and the model’s efficiency means typical running costs fall by 40%.

It’s a meaningful gap. For any organisation running large-scale automated tasks, that kind of reduction compounds quickly across millions of tokens.

Built for Long Tasks

The technical specification points clearly at extended, agentic work. Claude Opus 5.5 carries a default context window of one million tokens and a maximum output of 128,000 tokens, according to Anthropic’s API release notes. Adaptive thinking is enabled by default, meaning the model applies additional reasoning steps where it judges them useful, without the user having to configure it manually.

Anthropic positions the model squarely at complex coding, software development, extended document analysis, and knowledge work that unfolds over long sessions rather than single-turn exchanges. The one-million-token context window means it can, in principle, hold an entire large codebase or a large body of documents in a single session — a capability that matters considerably for automated development pipelines.

Who Can Use It

Access is available to Claude Pro, Max, Team, and Enterprise subscribers, according to Anthropic. Secondary reporting indicated that Anthropic raised five-hour usage limits for some plans at launch and that a temporary rate-limit reset for subscribers was available until 22 October 2026, though both details require confirmation against Anthropic’s current customer documentation.

Developers working outside consumer plans can access the model directly through the API. The model identifier for integration is `claude-opus-5-5`.

One point worth keeping in mind: lower per-token costs don’t automatically mean lower total bills. Organisations that run more tasks because the model is cheaper, or that use longer contexts more frequently, may find their overall spend holds steady or rises. The 40% figure is Anthropic’s estimate for typical workloads, and actual savings will depend on how the model is configured and used.

The Broader Picture

Claude Opus 5.5 enters a market where Anthropic is competing directly with OpenAI, Google DeepMind, and a growing number of open-weight model providers. The pricing move — chiefly the 40% typical-cost reduction — looks like a deliberate signal to enterprise and developer customers who are actively comparing running costs across providers.

No named individual from Anthropic was quoted in the launch materials available at the time of writing, though the company’s official announcement confirmed the model’s availability and pricing. Independent analysts have not yet published verified benchmark comparisons between Claude Opus 5.5 and the models Anthropic is comparing it against.

The performance equivalence claim — that Opus 5.5 matches Fable 5.1 on most work — is one to watch. If third-party evaluations bear it out, it would represent genuine value compression in Anthropic’s model lineup. If they don’t, the pricing advantage alone may still make the model attractive for cost-sensitive deployments.

What This Means for Kent Residents

Kent-based developers, businesses, and public-sector teams can access Claude Opus 5.5 through Anthropic’s consumer and enterprise plans or directly via the API, subject to plan eligibility. UK users will pay prices converted from Anthropic’s US-dollar list rates, so the actual cost in pounds will depend on exchange rates, applicable VAT, and the billing platform used. Any use of the model involving NHS, council, education, or personal data — for instance at NHS Kent and Medway ICB, Kent County Council, or Medway Council — would require separate data protection, information governance, and procurement assessments before deployment. For individual Kent residents on Claude Pro or Max subscriptions, the raised usage limits reported at launch may mean more headroom for extended tasks, though those limits should be verified against current Anthropic account documentation.