We’ve reduced cache read pricing by 75% on Claude Fable 5.1, making the model more cost effective. Here’s how admins can take advantage of these savings.
Available today on Claude Enterprise, Claude Fable 5.1 is our most advanced generally available model for coding and knowledge work, outperforming Claude Fable 5 across many benchmark evaluations, including CursorBench 3.2.0 (agentic coding), GDPVal-AA v2 (knowledge work), and Humanity’s Last Exam (multidisciplinary reasoning). Many of its largest gains are experienced in long-running agentic work, including multi-step business workflows, knowledge work that ends in a finished spreadsheet, deck, or memo, and coding tasks that run for hours.
We've cut cache read pricing for Claude Fable 5.1 by 75%, making Fable 5.1 more cost effective than Claude Fable 5. Our analysis suggests this will reduce the costs by 30-40% for Claude Code, 35-40% for Claude Design, and 15-25% for Claude Cowork for equivalent usage. For highly agentic workloads, the cost savings can be as high as 45% (Fig. 1).
Claude Fable 5.1 is also often more cost effective than Claude Opus 5 on long-running agentic tasks. Although its list token price is higher, it completes these tasks with less work (Fig. 2), and its cache read costs half as much.
Fig. 1: The same usage priced on Fable 5 vs Fable 5.1, by product surface and workload type; ranges reflect observed customer usage.
How to measure cost per task
Cost per task also depends on the effort setting you choose, in other words, how much reasoning the model does before it answers. More effort costs more per task, but often gets a better result. Claude Fable 5.1 defaults to high effort in Claude Code and medium in Claude Cowork and Chat. Effort caps, available in admin controls, let you set the ceiling for your organization and different ceilings for specific groups of users through role-based permissions.
While Claude Fable 5.1 has a higher cost per token than Claude Opus 5, its greater intelligence allows it to more efficiently accomplish complex agentic tasks and can make it the more affordable option. With Fable 5.1's cache reads now costing half of Opus 5's, we found it can provide greater intelligence for the same price and accomplish equivalent tasks for less. For example, on CursorBench 3.2.0, Fable 5.1 matches Opus 5’s best score at less than half the cost per task, and its top score exceeds Opus 5’s best score at any effort. Compared with Fable 5, Fable 5.1 delivers a higher score at a lower cost per task at every effort level, and at medium effort it matches Fable 5’s best result at approximately one fifth of the cost.
Fig. 2: CursorBench 3.2.0 scored against cost per task (log scale) at five effort levels.
Get started with Claude Fable 5.1 on Claude Enterprise
Like Claude Fable 5, Claude Fable 5.1 is a powerful model for enterprise organizations undertaking long-running, context-heavy work, including deep research, complex analysis, and code migrations.
It is available on Claude Enterprise today, and admins have three tools to manage access: model entitlements (who can access Claude Fable 5.1 in the model picker), effort caps (how much a task can spend), and defaults (what model and effort a new session starts on), at the org level, or for specific groups of users through role-based permissions.
Our recommended model configuration, by role:
Roles where unconstrained work is most of the job (e.g. deep research, complex analysis, and long-running agent work): make Claude Fable 5.1 their model default. This is the work it is best at, and with the new cache read pricing discount, is often the most cost effective model choice for this kind of work.
Roles where unconstrained work is part of the job: keep their current default, and offer Claude Fable 5.1 as an option through model entitlements so they can switch to it when a task calls for it.
Roles whose work is mostly constrained tasks (e.g. summarizing a document, drafting an email, extracting data from a report, answering routine questions): keep their current defaults on Claude Opus/Claude Sonnet.
How to manage model access in the admin console:
Model entitlements: Grant Claude Fable 5.1 org-wide or per role in Organization settings → Models. If you do nothing, Claude Fable 5.1 follows your Claude Fable 5 settings: where Claude Fable 5 is enabled today, Claude Fable 5.1 will be enabled automatically on September 8th, and you can opt out before then. If Claude Fable 5 is not enabled today, you must take action to grant access to Claude Fable 5.1, org-wide or per role. Until you do, your users cannot see or use Claude Fable 5.1, including the teams doing the work it is best at. Learn more: Manage model access for your organization.
Effort caps: Set a maximum effort per role, and, new in this release, a single maximum for the organization. Role settings cannot exceed the organization maximum. For cost-conscious organizations, Medium is the natural cap: at Medium effort, Fable 5.1 matches Fable 5's best results at about one fifth of the cost (Fig. 2). Learn more: Limit the maximum effort level for a custom role.
Defaults: Set the default model (org-wide or per role) and default effort (org-wide) so users start where you choose. If you do nothing, defaults set to Fable 5 move to Fable 5.1 on September 8th unless you opt out; everyone else keeps their current default and entitled users reach Claude Fable 5.1 only by switching to it by hand. Learn more: Set a default model for your organization.
If your team wasn’t able to access Claude Fable 5 due to data retention requirements, that may have changed today: Claude Enterprise customers who are eligible for Enterprise Frontier Safeguards (EFS) can access Claude Fable 5 and Claude Fable 5.1 through Claude Enterprise with their existing zero data retention setup.
Learn more about these cost reductions and Enterprise Frontier Safeguards by contacting your account team.