Fable 5.2 requires data retention by default; zero retention is limited to enterprises
Anthropic requires data retention by default for Fable 5.2 so its safety classifiers can run, and zero data retention is available only to eligible enterprise customers.
Anthropic's Fable 5.2 now requires data retention by default, and zero data retention is limited to eligible enterprise customers. According to the Claude Code account's post, as relayed in the Latent Space roundup, the retention requirement exists so Anthropic's safety classifiers can run.
The source doesn't say how long data is kept, what the classifiers inspect, or what the eligibility criteria are for enterprise accounts. The mechanism is easy to infer from the wording: classifiers that screen inputs or outputs need the content to be present somewhere.
The same launch also changed how subscribers reach the model. Ultra and Team Premium users get Fable "resets" scaled by tenure: one reset for accounts under six months old, two for six to twelve months, and three for accounts above twelve months, usable until March 8, according to @ClaudeDevs, as echoed by @bcherny and @edwinarbus. Pro and Team Standard users get a one-week trial capped at half the weekly limit, plus one reset. A separate limit caps Fable at 50% of a weekly allowance. @theo analyzed 1.1M logged requests on a $200 Ultra subscription and estimated the plan yields about $15,000 a month of Opus 5.5 usage but only about $3,000 of Fable 5.2 usage. Because Fable is capped at half the weekly limit, Theo wrote that Fable-only users "lose" half their limit. Those figures are his estimates from his own logs, not Anthropic's numbers.
Anthropic claimed Fable 5.2 is "a step above Opus 5.5 in almost all the capabilities we evaluated," and the roundup says it tops most independent leaderboards. Artificial Analysis's Intelligence Index v5 puts Fable 5.2 at 65 (max), ahead of Gemini 4 Carbon at 60 and Opus 5.5 at 59. Cost is less flattering. Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1. However, SemiAnalysis's GDPx benchmark measured cost per task rising from $67.16 to $88.15, a 31% increase. The run logs point to the extra checking: Fable writes hundreds or thousands of test cases where Opus 5.5 and Fable 5.1 wrote about 20, and spawns subagents to check its work. SemiAnalysis attributes its first-place 16.4% GDPx score to that testing breadth.
The effort settings complicate any cost comparison. Anthropic reports that Fable at High matched Opus 5.5 at Max for about half the cost per task on FrontierCode 2.0. Artificial Analysis found that at max, Fable 5.2 costs about $26.5 per task, roughly 2.6 times Opus 5.5's $10.0. Vals found High to be the best-performing effort level, above xhigh and max. @nrehiew_ noted that only the Fable models show that max-reasoning dip, with Opus and Sonnet not showing it. The roundup records @Yuchenj_UW's blunt advice, "Don't use Fable 5.2 Max," and @theo's reading that "don't use max reasoning" is an alternate reading of Anthropic's own recommendation.
Several commentators framed the launch in terms of Anthropic's pending IPO, though that is opinion. @andersonbcdefg wrote that he was "not sure this model is good enough to save the IPO," and SemiAnalysis described the release as "a marginal improvement, incomparable to the original Mythos release, and likely far behind what Anthropic has internally." The roundup also notes that @TheRundownAI described Fable 5.2 as a "smaller jump over Opus 5.5," after larger gains over Fable 5.1.
What changed
Fable 5.2 now requires data retention by default, and zero data retention is limited to eligible enterprise customers.
Type: Breaking change · Applies to: Fable 5.2
Why it matters
Teams with strict data handling rules may not be able to use Fable 5.2 unless they qualify for enterprise zero-retention terms.
What you can do
Check your data retention requirements against Fable 5.2 terms, and ask Anthropic about zero-retention eligibility before using it on sensitive data.
Original source
We have no link to the original source for this story yet.
Where we saw it
Latent Space
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works