Anthropic Ships Claude Fable 5.2 and Mythos 5.2: 20% Cheaper API, a Free Week for Pro, and a 30-Day Retention Catch

This is an analysis of public material: Anthropic's announcement, developer docs and pricing page, Help Center articles, the AWS launch post, and press coverage. We did not run the model or reproduce any result; every benchmark and anecdote below is Anthropic's own claim or a reporter's account.

Who should care

• Claude Pro and Team standard-seat subscribers: Fable normally runs on pay-as-you-go usage credits on these plans. Through November 16 at 11:59 pm PT it is included in your plan, capped at 50% of weekly limits.
• Max, Ultra and Team premium-seat subscribers: Fable was already included; you now also get one to three "Fable resets" depending on tenure.
• Teams paying API bills: anyone already on Fable 5.1 gets a 20% lower rate by switching the model ID. The bigger decision may be whether you need Fable at all versus Opus 5.5.
• Security and life-sciences organizations: Mythos 5.2 requires approval through the Cyber Verification Program or Life Sciences Verification Program.
• Enterprise admins: Fable 5.2 is off by default on Enterprise plans, but is added automatically to Fable-enabled roles seven days after launch unless you opt out.

What Anthropic has confirmed

The prices come straight from Anthropic's pricing page, which still lists Fable 5.1 at $10/$50, and the specs from the models overview. Notably, Anthropic itself does not position Fable as the default: it tells users to start most coding and knowledge work on Claude Opus 5.5 ($4/$20) and move up to Fable or Mythos only for the most ambitious, open-ended technical problems. It also pitches Fable 5.2 as an "Advisor" that smaller models consult, or an "Orchestrator" that hands simpler subtasks to cheaper models.

The headline claims

• Code that gets merged: on FrontierCode 2.0, which scores whether maintainers would accept a model's changes, Anthropic reports 49.1% at Xhigh effort for Fable 5.2, versus 32.4% for Fable 5.1, 45.5% for Opus 5.5 and 42.8% for OpenAI's GPT-6 Astra. At High effort, Fable 5.2 roughly matched Opus 5.5's best score at about half the cost per task, the company says.
• Multi-day agent runs: an Anthropic engineer gave Fable 5.2 an empty folder and asked for a prototype operating system; the model coordinated more than 1,000 subagents for eight days and produced a system that boots on a laptop. A separate 1,000-agent run formalized Fermat's Last Theorem in Lean in 30 hours.
• Security (Mythos 5.2): on Anthropic's internal CVE Rediscovery test across 1,829 published CVEs, Mythos 5.2 scored 89.6% versus 78.3% for Mythos 5.1, and found 10.2% more vulnerabilities than Opus 5.5. A Firefox security tester quoted in the post said Mythos 5.2 roughly doubled its vulnerability-discovery metrics compared with Mythos 5.1.
• Fewer false refusals: Anthropic says Fable 5.2's cyber and bio safeguards are as robust as 5.1's while blocking fewer legitimate requests.

What is not yet established

• Independent replication: none of the benchmarks or showcase projects above has been reproduced by a third party in the sources reviewed for this article. Korean outlets reported that Fable 5.2 placed first on Artificial Analysis's Intelligence Index at 65 points; we did not find that figure in a primary source and treat it as reported, not confirmed.
• Mythos expansion abroad: Korean coverage (Aju Business Daily) says Anthropic plans to open Mythos to companies outside the US; Anthropic's own post only says it is "continuing to expand" the Cyber Verification Program. No timeline is public.
• Export-control risk: in June, a US Commerce Department order took Fable 5 and Mythos 5 offline worldwide for 19 days before Anthropic redeployed them on June 30. Nothing similar has been announced for 5.2, but the episode is a reminder that a single frontier model can disappear from a production stack overnight.
• Real-world cost: Fable consumes plan limits faster than other models, and Anthropic has not published typical token counts per task outside its own benchmark charts.

What it means for you

For individual subscribers, the practical news is the launch offer. Pro and Team standard seats can spend up to half their weekly limit on any Fable model (5.2, 5.1 or 5) until November 16 at 11:59 pm PT (2:59 am ET on November 17), plus one reset that expires at the same moment. Max, Ultra and premium seats get one reset if they have subscribed for under six months, two for six to twelve months and three beyond that, all expiring March 8, 2027 at 11:59 pm PT. Resets only go to accounts whose paid billing was in place before 10:00 am PT on November 9. The Free plan, self-serve Enterprise and usage-based Enterprise plans are excluded. Your default model does not change, so you have to pick Fable 5.2 from the model picker, and Claude Code needs version 2.1.321 or later.

For engineering and compliance teams, the retention terms matter more than the discount. The developer docs classify both models as Covered Models that require 30-day retention, which rules out existing zero-data-retention arrangements unless Anthropic signs off. According to the AWS launch post, Bedrock users must set a data-retention mode on the account or project before invoking the model, with Amazon staff performing safety review inside the AWS boundary; eligible customers can use zero retention for internal workloads until Anthropic's Enterprise Frontier Safeguards arrive. Teams in regulated industries such as healthcare or finance should check those terms against their own obligations before routing sensitive data to Fable 5.2.

Limits of this article

This piece relies on Anthropic's announcement, developer documentation, pricing page, Help Center articles and Supported Regions Policy, the AWS Machine Learning blog, and reporting by Reuters (via The Star), Yahoo Finance and Aju Business Daily, all checked on November 11, 2026. We had no access to the model and reproduced nothing. Prices, promotions and availability can change, and the system card's detailed safety findings are outside this article's scope.

Last verified 2026-11-11.