Claude Fable 5: Pricing and Data Retention

Claude Fable 5Anthropic's highest-priced generally available Cursor model for long, complex work, with a 30-day retention opt-in for eligible privacy and enterprise accounts. Press Enter for the full definition. is an Anthropic Mythos-class model built for long-running, complex, and asynchronous work. Cursor places it at the top of CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition. alongside Opus 5 and rates it medium for speed, high for cost, and frontier for intelligence. Its stated advantage is holding intent across a long task with fewer check-ins than earlier models needed.

The price and retention terms make Fable an escalation model rather than a casual default. Cursor says it costs about twice as much as Opus 5 and recommends it for a team's most complex work. Use that positioning literally: choose a task with a costly handoff or a long autonomous run, define the expected evidence, and compare the completed result against Opus before standardizing.

Fable 5 has a specific retention exception. Cursor states that Anthropic stores agent input and output data for automated and human harm-prevention processes regardless of Cursor Privacy ModeCursor's setting that guarantees code data is not used for training by Cursor or its model providers, and that an admin can enforce org-wide; data-retention terms are a separate, contractual layer. Press Enter for the full definition.. Retained customer data is automatically and permanently deleted after 30 days unless a safety investigation or legal hold applies. It is not readable by a person by default and is not used to train Anthropic models or products.

Cursor also says retained enterprise data is not used to train Claude and is not shared with other customers. Enabling Fable does not change the account's Privacy ModeCursor's setting that guarantees code data is not used for training by Cursor or its model providers, and that an admin can enforce org-wide; data-retention terms are a separate, contractual layer. Press Enter for the full definition. or Cursor's own retention behavior. Those distinctions matter: Privacy Mode's no-training commitment remains in place, while this model adds provider retention for the stated safety purpose.

A 30-day deletion statement is not the same as zero data retention. If a repository or customer agreement permits only zero-retention providers, Fable does not fit that policy even though training is prohibited. Keep the model disabled for that scope and choose an allowed alternative instead of trying to remove sensitive strings through prompting alone.

Customers with Privacy ModeCursor's setting that guarantees code data is not used for training by Cursor or its model providers, and that an admin can enforce org-wide; data-retention terms are a separate, contractual layer. Press Enter for the full definition. enabled must approve Fable 5's Data Retention Policy in the Cursor Dashboard before the model appears. On Teams, that approval applies to the whole team. On Enterprise, availability is approved at the organization level, then team and group model access controls continue to decide who can select it.

Fable is off by default for every team with Privacy ModeCursor's setting that guarantees code data is not used for training by Cursor or its model providers, and that an admin can enforce org-wide; data-retention terms are a separate, contractual layer. Press Enter for the full definition. enabled and for every Enterprise customer, regardless of the Enterprise account's Privacy Mode setting. Individual customers and Teams with Privacy Mode disabled get it on by default. That default is a product availability rule, not a statement that the retention terms fit a company's policy.

1. 1Read the current Fable data-retention policy and map it to the data allowed in the target repositories.
2. 2Approve the policy at the required account, team, or organization level in Cursor Dashboard.
3. 3For Enterprise, restrict model access to the intended teams or groups after organization approval.
4. 4Confirm the model appears in a fresh Cursor session and run a non-sensitive test task.
5. 5Record the approver, policy version or review date, scope, and next review date.

Team-wide acceptance deserves a named owner because one approval changes availability for more than the person clicking it. Enterprise group controls can narrow use afterward, but the organization has still accepted the provider terms. Review that decision again when Anthropic or Cursor changes the policy, rather than letting an old dashboard state stand in for current legal approval.

Cursor documents an automatic fallback when a request trips Fable 5's security guardrails. The request is routed to Claude Opus so the work can continue without a manual retry or model switch. This means a session selected as Fable may contain work produced by Opus after the guardrail event.

Treat the fallback as continuity, not proof that both models share identical behavior, rates, or approval status. If model identity matters for an audit, cost model, or evaluation, inspect the usage record and document what actually ran. Cursor's model page explains the fallback behavior but does not turn it into a guarantee that every governed workflow can ignore the model change.

Fable 5 uses the third-party Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. pool. Cursor lists $10 per million input tokens, $12.50 cache write, $1 cache read, and $50 output. Those rates are twice the corresponding Opus 5 rows. Cursor recommends the high thinking variant for the strongest results, which can also increase the amount of work and tokens spent on a difficult request.

The useful unit is cost per verified task, not the cheapest token row. A long Fable run can be economical if it completes work that would take several Opus retries and human interventions. It is expensive when used for a small change that ComposerCursor's own fast coding model, tuned for the editor and priced well below frontier models; the recommended day-to-day model for executing a plan. Press Enter for the full definition. or Sonnet can pass in one attempt. Evaluate both outcome and supervision time on your own task set.

Use Fable for a long, asynchronous task with a clear finish line and data that is allowed under the retention policy. Good candidates include a difficult cross-system migration, a broad investigation that must keep a constraint set intact, or a background implementation where repeated human check-ins are the main cost. Give the run tests, review criteria, and a stopping condition before it starts.

Choose Opus, Sonnet, Grok, or ComposerCursor's own fast coding model, tuned for the editor and priced well below frontier models; the recommended day-to-day model for executing a plan. Press Enter for the full definition. when the policy, price, or task shape points elsewhere. The most capable model in one benchmark is not automatically the safest or most efficient default for a team. Fable's useful niche is narrow but real: hard work that benefits from sustained intent and justifies both the retention exception and the published premium.