On 9 November 2026, Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2. According to Anthropic’s announcement, the two are the same underlying model: Fable has stronger cybersecurity and biology safeguards and is available to all users, while Mythos is available only to vetted organizations in dual-use fields. Anthropic noted on the page that it was amended on 11 November to clarify the formalization methodology. Trade press including SiliconANGLE and SD Times reported the launch on the same day.
What was announced
- Availability. Anthropic says Fable 5.2 is available on all platforms, including AWS, Google Cloud and Microsoft Azure, under the API model ID
claude-fable-5-2. - Mythos access. Cybersecurity professionals apply through the Cyber Verification Program; life sciences researchers apply through the Life Sciences Verification Program. Mythos fast mode is limited to organizations in those programs.
- Pricing. Anthropic lists $8 per million input tokens and $40 per million output tokens as the standard rate, and an optional fast mode at $16 and $80 that it says speeds up generation by up to 2.5 times. It says output is up to 30% faster than Fable 5.1 and Mythos 5.1.
- Positioning. Anthropic describes Fable and Mythos as best for the most ambitious, open-ended technical work, and continues to recommend Claude Opus 5.5 as its everyday model.
Capability claims (Anthropic’s own)
The announcement includes a benchmark table comparing Fable 5.2 with Fable 5.1, Opus 5.5 and competing models from other developers. Examples Anthropic reports include 49.1% on FrontierCode 2.0 (against 32.4% for Fable 5.1), 74.6% on Terminal-Bench-Science 0.1 (75.3% for Mythos 5.2, against 52.6% for Fable 5.1) and 74.2% on Humanity’s Last Exam with tools. Anthropic notes that some scores use each lab’s own unique tasks and that some were run internally.
For science, Anthropic reports that about 1,000 Fable 5.2 agents completed a formalization of Fermat’s Last Theorem in 30 hours, compared with 11 days for an earlier Anthropic research version; that Fable 5.2 solved two open black-hole problems using holographic duality; and that Mythos 5.2 accurately modeled 28 of 34 attempted AAV capsid structures using two open-source protein folding models. These are developer-reported results described in a launch announcement, not peer-reviewed findings.
Safeguards and safety testing
Anthropic says its cyber and biology classifiers block high-risk cyber activity and most research biology on Fable 5.2, with blocked cyber queries routed to Opus 4.8 and blocked biology queries to Opus 5. It reports about 55% fewer false-positive blocks than Fable 5.1. It also says training reduced capabilities in accelerator kernel optimization and LLM architecture improvement to roughly Opus 5.5 levels.
The UK AI Security Institute, Frontier Design and METR evaluated the model before release, according to Anthropic; the system card summarizes their reports, and Anthropic states they do not endorse the blog’s claims. Anthropic assesses the risk of high-stakes unsafe behavior as low, and also states that both models voiced suspicion of being evaluated more often than recent models, which it says makes evaluations harder to interpret. Details are in the system card.
Data handling
Anthropic says some misuse patterns only appear across multiple requests, so its safeguards monitor a window of usage data. It states that eligible customers can use both models with zero data retention, while other customers have 30-day retention for safety monitoring. It also describes Enterprise Frontier Safeguards, which run safety monitoring on prompts and responses stored in the customer’s own infrastructure and which it says is in private beta.
What it means for researchers and research offices
- Access differs by field. Because Fable 5.2 blocks most research biology queries, life sciences groups may find general access limited; Anthropic’s route for them is the Life Sciences Verification Program, which involves an application and organizational vetting. Any use of verified access is an institutional decision that may warrant review by research security and compliance staff.
- Retention terms matter for sensitive data. Zero data retention is described as available to eligible customers only; the default is 30-day retention for safety monitoring. Offices setting rules for human-subjects, export-controlled or unpublished data should check which terms apply to their contract.
- Treat benchmark results as vendor claims. The figures above were published by the developer, several on internal or developer-chosen tasks. Independent replication, and the external evaluators’ own reports, are the better basis for procurement or policy decisions.
- Disclosure obligations are unchanged. Journal and funder policies on disclosing generative AI use apply regardless of model version; see CASRAI’s earlier coverage of AI disclosure and Claude Fable 5.
- Release cadence. Frontier model versions are changing quickly, and institutional policies that name specific model versions will age quickly; see AI model release cadence and institutional policy review and the earlier report on export-control action affecting Fable 5 and Mythos 5.
Sources
- Anthropic, Claude Fable 5.2 and Mythos 5.2 announcement (9 November 2026, amended 11 November)
- Anthropic, Claude Fable 5.2 and Mythos 5.2 system card
- SiliconANGLE, 9 November 2026
- SD Times, 9 November 2026
