Anthropic release announced Monday, November 9, 2026; coverage November 10, 2026
Same Model, Two Doors: Fable 5.2 for Everyone, Mythos 5.2 Only for Vetted Cyber and Life Science Teams
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, two configurations of the same underlying model separated by safeguards and by who may use them. Fable 5.2 is broadly available on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. Mythos 5.2, described by Anthropic as its most capable model for cybersecurity engineering, is restricted to organizations vetted through its Cyber and Life Sciences verification programs, because it retains capabilities in cybersecurity and biology that Fable blocks. Both are priced at 8 dollars per million input tokens and 40 dollars per million output tokens, down from 10 and 50 dollars in the prior generation, a 20 percent cut, with output up to 30 percent faster. The capability claims are the company's own and are labeled as such here. On FrontierCode 2.0, Fable 5.2 scored 49.1 percent at its highest effort setting against 32.4 percent for Fable 5.1, and its high effort setting matched Opus 5.5 at its highest setting at roughly half the cost per task. On Irregular CyScenarioBench, which measures planning and executing a multi-stage cyber operation in a realistic environment, Fable 5.2 completed 76.3 percent of challenges, against 61.7 percent for Fable 5.1 and 61 percent for OpenAI GPT-6 Cyber. On the CVE Rediscovery evaluation, covering 1,829 open-source advisories from October 2026, the 5.2 models identified 89.6 percent of vulnerabilities, up from 78.3 percent for Mythos 5.1, with better reproduction and fixing reported as well. Anthropic says that during pre-release work the models completed otherwise infeasible software migrations, designed three working electronic devices, solved two open problems in black hole physics and formalized the proof of Fermat's Last Theorem in 30 hours, and that in one internal experiment Fable 5.2 coordinated more than 1,000 subagents over eight days to build a prototype operating system that booted on a laptop. Its system card says high-stakes misalignment risks remain low and that the models were roughly an order of magnitude less likely than Opus 5.5 and Sonnet 5.5 to publish harmful content to public package registries without encouragement, while still doing so when encouraged. Two terms will matter more to regulated buyers than any benchmark: customers eligible for zero data retention keep it, while other customers face 30-day retention for safety monitoring, and a private beta called Enterprise Frontier Safeguards moves automated safety monitoring onto prompts and responses stored inside customer infrastructure. Claude Enterprise customers can also reach the Mythos vulnerability detection capability through Claude Security, which Anthropic updated with continuous code scanning and automatic fix tracking, without joining the verification programs.
Executives
The strategic fact is the two-door design. The same capability now ships in a general version and a gated version, and the gate is the vendor verification program, which means your access to the strongest defensive tooling depends on a vendor decision about your organization. Note also what the price cut does not settle: analysts covering the launch caution that effort settings, not token prices, drive real cost per completed task, so evaluate on finished work. Before any regulated workload moves to these models, get the retention position in writing, because the gap between zero retention and a 30-day safety monitoring window is a compliance fact, not a pricing detail.
CIO / CISO
Plan the defensive use and the exposure in the same memo. Defensively, the CVE Rediscovery and fix capability, reachable through Claude Security for Enterprise customers, is a concrete backlog tool: point it at your own code first and measure it against your scanner findings. On exposure, a model completing 76.3 percent of a multi-stage cyber operation benchmark is available in gated form to vetted organizations and in reduced form to everyone; assume attacker capability rises on the same curve and that your detection timelines should shorten accordingly. Confirm which of your existing Claude agreements carry zero data retention before enabling 5.2 for sensitive work, and assess the Frontier Safeguards beta if retention is your blocker.
Engineers
Benchmark on your own work before adopting the headline numbers. FrontierCode and CyScenarioBench are vendor-reported evaluations with vendor-chosen settings, and the pre-release feats, the operating system build, the theorem proof, are Anthropic accounts of internal experiments, impressive and unverified in equal measure. Run Fable 5.2 against your migration and review tasks at the effort setting you would actually pay for, log cost per accepted change, and compare against your current model on the same set. If you test the security tooling, start with repositories whose known findings give you a ground truth to score against.
Everyone (non-IT)
The plain version: the newest Claude models are faster and cheaper, one is open to all users, and the strongest version for security work is deliberately kept behind an approval process because it is also the most capable of finding ways into software. Nothing about your everyday use needs to change. If your employer announces access to the new models, expect new rules about what data may be pasted into them, because the retention terms differ by contract, and those rules, not the model name, are what protect your work.