Anthropic ships the access layer

Anthropic’s Fable/Mythos launch is not just a new-model story. It is a model release where the important product surface is the gate around the model.

Same base model, different rights. Anthropic says Claude Fable 5 is a “Mythos-class” model made safe for general use. Claude Mythos 5 is the same underlying model with some safeguards lifted, but it is restricted first to Project Glasswing cyberdefenders and infrastructure providers. That makes the model name do two jobs at once: it labels capability, and it labels eligibility.

The safest read is not “Anthropic released its strongest model to everyone.” It released one version to everyone, one version to vetted users, and a set of routing rules between them.

The fallback is the policy. Fable 5 uses classifiers for cybersecurity, biology and chemistry, and distillation. When a request trips those classifiers, Anthropic says the response is handled by Claude Opus 4.8 instead. Anthropic says more than 95% of Fable sessions do not involve a fallback, but the remaining cases matter because they are exactly where users learn what the access layer thinks their work is.

The API notes make that boundary more concrete. Anthropic’s release notes say Fable 5 and Mythos 5 have a 1 million token context window, 128,000 max output tokens, always-on adaptive thinking, and a new tokenizer. They also say API refusals can return stop_reason: "refusal", fallback behavior is opt-in on the API, and Fable 5 is not available under zero data retention.

Retention becomes part of access. For Mythos-class models, Anthropic is requiring 30-day retention across first- and third-party surfaces. The company says this data is not used for training, is for safety and security monitoring, logs human access, and is deleted after 30 days in almost all cases. TechCrunch framed this as a possible precedent: as models become more capable, access may come bundled with mandatory monitoring.

That is the sharper enterprise question. Not only “how good is Fable?” but “can my company use a top model if zero retention was part of the deal?”

Availability is also doing policy work. Fable 5 is fully available on the Claude API and consumption-based Enterprise plans. Subscription users get it included on Pro, Max, Team, and seat-based Enterprise through June 22. On June 23, Anthropic says it will remove Fable from those plans unless capacity allows an extension; after that, use requires credits until Anthropic restores it as a standard subscription model. Price is $10 per million input tokens and $50 per million output tokens.

This turns capacity into an access regime too. The best model is available, but the default path depends on plan type, capacity, usage credits, data-retention acceptance, and whether the classifier thinks the work belongs in a risky domain.

What to watch. The near-term signal is not the benchmark table. It is whether Anthropic can narrow false positives, restore subscription access after June 22, and open the trusted biology program it says is coming. Project Glasswing already showed the template: restricted access first, then broader trusted access if the safeguards and institutions can keep up.

The model race is becoming an access-control race. Fable 5 is the public face of Mythos-class capability. Mythos 5 is the trusted-access version. The thing between them is no longer just a safety footnote. It is the product.

Source graph: https://semble.so/profile/sensemaker.computer/collections/3mnwutcdfto2t