What happened
Cursor’s documentation page for Claude Fable 5 says Cursor recommends Claude Fable 5.2, describing it as Anthropic’s latest Mythos-class model and saying it has lower rates than Fable 5. The same page labels the model ID claude-fable-5; it lists a 300,000-token context window, a maximum context of 1 million tokens, Anthropic as provider, and “Frontier” intelligence. The documentation does not spell out whether every listed capability or price applies specifically to version 5.2 or to the model page’s Fable 5 entry.
Cursor describes Fable 5 as its most capable generally available model for autonomous knowledge work and coding. The company says it is intended for long-running, complex, asynchronous tasks that previously required more frequent check-ins. It says Fable 5 tops CursorBench alongside Opus 5, and costs about twice as much as Claude Opus 5. A thinking variant is available; Cursor recommends the high thinking variant for the strongest results. These are Cursor’s descriptions and benchmark claims, not an independent evaluation.
Availability depends on account settings. Cursor says the model is on by default for individual customers and Team accounts without Cursor Privacy Mode. It is off by default for every Enterprise customer and for Teams with Privacy Mode enabled, and those users require an administrator to enable it. For accounts using Privacy Mode, the administrator must approve Anthropic’s Data Retention Policy in the Cursor Dashboard. Team approval applies across the team; Enterprise availability is approved at the organization level, while team and group controls can further restrict access. Cursor says using Fable does not change its own Privacy Mode settings or Cursor’s own retention of data.
When Fable is used, Anthropic stores agent inputs and outputs for automated and human harm-prevention processes, regardless of Cursor Privacy Mode. Cursor’s page says the data is not used to train or improve Anthropic models or products. It says retained customer data is automatically and permanently deleted after 30 days unless subject to a safety investigation or legal hold, is not readable by a person by default, and is not shared with other customers. The page says retained enterprise data is not used to train Claude.
The model is billed separately from usage included with Cursor plans. Cursor lists rates per million tokens: $10 for input, $12.50 for cache writes, $1 for cache reads and $50 for output. On the X Subscription, charges come from prepaid credit; on legacy Cursor plans, usage draws from the Other Models pool. Fable has access to Cursor agent tools, including file search and reading, web search, editing, shell commands, browser control, image generation, background questions and rules retrieval. Cursor says that if a request triggers Fable’s security guardrails, it routes the request to Claude Opus automatically rather than returning an error.
What the archive adds
New Runtime’s earlier coverage on Nov. 9 described Claude Fable 5.2 as a new Anthropic model and noted a 30-hour formalization of Fermat’s Last Theorem. That earlier note concerned the model’s announcement and research demonstration; Cursor’s documentation adds a distinct distribution and operational layer: access controls, metering and retention terms inside a coding tool. This is a new fact on a known model story, not a restatement.
Trend signal
This supports the assumption that developers will continue using several AI clients, including Cursor and Claude, on the same project rather than standardizing on one vendor’s integrated memory. New Runtime’s shared-context infrastructure shift describes that multi-client direction. Cursor’s documentation does not establish shared memory or interoperability; it is evidence of a model becoming selectable in another client, not of a common context layer.
Why now
Cursor’s current documentation recommends Fable 5.2 while retaining a page centered on the Fable 5 model ID and terms. That makes the practical availability details relevant to teams considering the newer model, but the page does not provide a dated change log entry establishing when the recommendation or availability changed.
Why it matters
Teams evaluating Fable in Cursor need to account for more than a model selector. Separate token billing can make usage materially different from models included in a plan, while the retention opt-in can require an organization-wide decision. The stated 30-day deletion policy and non-training commitment may help a team assess suitability, but they do not remove the need to check internal data rules and the exact Anthropic terms before enabling access.
Cursor’s automatic fallback also changes how teams should evaluate guarded requests: a task may continue on Claude Opus rather than fail, but that means the model handling a portion of work can differ from the one selected. Teams should test representative tasks, record when fallback occurs, and compare quality and total token costs before making Fable a default for a workflow. Cursor’s own benchmark claim is a useful signal to investigate, not a substitute for that workload-specific test.
Limits and uncertainty
Cursor’s page does not clearly distinguish which listed specifications, tool availability, retention terms and prices apply specifically to Fable 5.2 versus the Fable 5 model ID. It provides no date for the recommendation and no independent benchmark results or methodology on this page. Actual cost will depend on token usage, including cache behavior, and the page gives no typical task cost. Cursor does not quantify how often guardrails trigger fallback to Opus or how that affects outcomes. Teams should confirm the current model identity and terms in their account before relying on the listed details.
Skill to build now
Run a small, representative evaluation before enabling Fable team-wide. Choose a few multi-step coding tasks, define what counts as a correct completion, and record model selection, fallback, input and output tokens, and review time. Compare results with the model your team uses today. Separately have the account or organization administrator review Anthropic’s retention terms and decide which groups, if any, should receive access.
Reader takeaway
Cursor recommends Claude Fable 5.2, but the documentation’s detailed terms are attached to a page identified as claude-fable-5. Treat the listed price, retention conditions and fallback as operational factors to verify—not just fine print—before putting the model into a team workflow.