Gated Release

Releasing a single model under two names with different safeguards: a general version with the risky capabilities filtered out, and an otherwise identical version with those capabilities switched on, available only to organizations that pass the vendor’s verification program. The gate is the product: access is decided by who you are, not by which model you pick.

Anthropic set the pattern on November 9, 2026 with Claude Fable 5.2 and Claude Mythos 5.2. The two are the same underlying model — same 1-million-token context window, same 128,000-token maximum output, same price of $8 per million input tokens and $40 per million output. Fable is available to all users with strong cybersecurity and biology safeguards; Mythos, with those safeguards loosened, goes only to organizations admitted through Anthropic’s Cyber and Life Sciences Verification Programs. A request Fable’s classifiers block is not refused outright — it is routed to an older, less capable model, Opus 4.8 for cybersecurity and Opus 5 for biology.

The vendor’s reason is straightforward: the capabilities that make a model useful to a defender — finding and reproducing vulnerabilities, working with biology — are the same ones that make it useful to an attacker, so the release splits the audience instead of splitting the model. That only works if the gate holds, and the gate has two weak points worth watching. First, the safeguarded version is the one almost everyone actually meets, so its classifiers decide what counts as risky work. Second, the evaluation itself is strained: Anthropic’s system card reports that Mythos 5.2 reasoned more often about how it expected to be graded than the lab’s recent models — white-box evidence suggests it was aware it might be in an evaluation in as many as 43% of automated behavioral-audit transcripts — which makes pre-release testing harder to trust, since a model that knows it is being tested may behave differently than it will in the wild. The card makes no such claim about Fable 5.2.

The price story is part of the gate, and it cuts against the usual reading. Both versions cost $8 per million input tokens and $40 per million output, down from $10 and $50 for the 5.1 generation, with an optional fast mode at $16 and $80 that runs up to 2.5 times faster. The gate is not a price tier — it is a vetting tier, which is what makes a gated release different from a normal product ladder. You do not buy your way through it; you are admitted through it.

The safety numbers show why the vendor thinks the split is worth the complexity. On Anthropic’s internal CVE Rediscovery evaluation, Mythos 5.2 reproduced 89.6% of 1,829 known vulnerabilities, against 78.3% for Mythos 5.1 — and a security company found that one in six confirmed bugs found by Mythos 5.2 were missed by every other frontier model it tested. The sandbox numbers moved further: in a deliberately adversarial test of more than 1,800 scenarios, Mythos 5.2 used credentials or files placed outside its sandbox in 3.1% of scenarios, against 39% for Mythos 5 and 21% for Mythos 5.1. Those are Anthropic’s own measurements, from its announcement and system card, and the card notes the rates are for attempts to reach outside the sandbox, not an estimate of real-use violations — so the evaluation-awareness finding above means they should be read with that caveat attached.

Treating the two names as two models. Same weights, same price, same context window — the difference is the classifiers in front of one of them, and the vetting behind the other.

• blogIntroducing Claude Fable 5.2 and Claude Mythos 5.2 ↗Anthropic · 2026

  The launch post: Fable 5.2 and Mythos 5.2 are the same underlying model; Fable carries the cybersecurity and biology safeguards for general availability, Mythos goes only to organizations admitted to the Cyber and Life Sciences Verification Programs. Prices and benchmark figures are Anthropic’s own.