Claude Fable 5.1 is putting a lower price on cached context rather than cutting its headline API rates. Anthropic lists the model at $10 per million input tokens and $50 per million output tokens, the same input and output prices shown for Fable 5. Cache reads now cost $0.25 per million tokens, which Anthropic says is 75% lower than the earlier model.

The difference matters for long-running AI agents. An agent may read the same instructions, files or project history across several steps. If that material is stored in a prompt cache, the application can pay the cache-read rate instead of paying the full input price every time. The saving depends on how much context can be reused and how the workload is designed.

Anthropic estimates that the change can reduce the cost of typical workloads by about 25%. For highly agentic work, the company says the saving can reach approximately 45%. Those are Anthropic’s estimates, not an independent benchmark or a promise for every API bill. Developers still need to measure cache hits, input volume, output volume and failed requests on their own systems.

The model is built for demanding coding, knowledge work and long-horizon tasks. Anthropic’s platform documentation lists a one-million-token context window, a maximum output of 128,000 tokens and adaptive thinking that is always on. The model was released on September 1, 2026, and the documentation marks it as an active model with a reliable knowledge cutoff in June 2026.

A large context window does not make an AI system a current information source. The cutoff means Fable 5.1 may not know about later events unless an application supplies updated material through retrieval or another tool. Anthropic’s own model documentation also presents the model as a system for coding and knowledge work, not as a substitute for checking primary records.

Access is spread across more than one route. Anthropic says Claude Fable 5.1 is available on its Claude Platform and to eligible Pro, Max, Team and Enterprise users. Developers can also use it through Amazon Web Services, Google Cloud and Microsoft Foundry. The exact controls, billing relationship and data terms depend on the platform chosen.

Anthropic is also highlighting enterprise privacy controls. Its Claude Fable page says customers eligible for Enterprise Frontier Safeguards can store data on their own cloud infrastructure, with human review handled by the customer by default. That option is separate from the cheaper cache-read rate and should not be treated as a universal zero-data-retention promise.

The Claude Fable 5.1 story is therefore as much about operating cost as model quality. Anthropic is keeping the visible input and output rates while making repeated context cheaper for suitable workflows. Teams considering the model should test real prompt reuse, verify platform terms and keep a human review path for high-impact work before turning the announced estimates into a budget.