What Is Claude Mythos 5? The Model Behind Fable 5

When Anthropic announced Claude Fable 5 and Claude Mythos 5 on June 9, 2026, most of the attention went to Fable - the model anyone can use. But the more unusual story is Mythos 5: the same underlying model, shipped under a different name, with its cybersecurity safeguards lifted, and available only to a vetted group of government and critical-infrastructure partners. This explainer covers what Mythos 5 is, where it came from, and why Anthropic released a model it once considered too dangerous to ship.

Mythos: a model class above Opus

For years, Anthropic's lineup topped out at Opus, with Sonnet and Haiku below it. Mythos is a new tier above Opus - Anthropic's designation for its most capable model class. Anthropic describes Fable 5 as a "Mythos-class model made safe for general use," and says plainly that "Fable's capabilities exceed those of any model we've ever made generally available."

The naming signals the relationship: Mythos is the raw frontier class; Fable is the public-facing configuration of it.

Mythos Preview: the model Anthropic withheld

Mythos 5's predecessor, Claude Mythos Preview, never reached the public at all. Completed in April 2026, it demonstrated what Anthropic characterized as superhuman cyber-exploit capabilities - and the company concluded that no organization, itself included, had safeguards strong enough to release it generally. Instead, roughly 50 vetted partners received access.

The decision wasn't purely defensive caution. Put to work on defense, Mythos Preview found more than 10,000 high- and critical-severity zero-day vulnerabilities, including flaws in every major operating system - a concrete preview of what a model of this class can do for security research when pointed in the right direction.

Project Glasswing: who actually gets Mythos 5

Access to Mythos-class models runs through Project Glasswing, Anthropic's cyberdefense program operated with the US government and critical-infrastructure operators. Glasswing partners - the same small circle, on the order of 50 organizations, that worked with Mythos Preview - use the model for defensive security work: vulnerability discovery, patching, and hardening of essential systems.

Anthropic calls Mythos 5 the model with "the strongest cybersecurity capabilities of any model in the world." Independent benchmark data is consistent with the claim: on ExploitBench, Fable 5 - the restricted configuration - scores 78.0 against Opus 4.8's 40.0 and GPT-5.5's 34.0. Mythos 5 is that capability without the cyber classifier in the way.

One model, two safety configurations

The key technical fact: Fable 5 and Mythos 5 are the same underlying model. The difference is the safety system wrapped around it.

Fable 5 (claude-fable-5) ships under ASL-3 and CB-1 protections, with three real-time classifiers - cyber, bio/chem, and reasoning-extraction/anti-distillation - plus undisclosed interventions limiting help with frontier LLM development. When a session trips the safeguards, it can fall back to Opus 4.8, which Anthropic says happens in under 5% of sessions.
Mythos 5 (claude-mythos-5) is the same model with the cyber safeguards lifted, for partners whose work those safeguards would otherwise block.

Pricing is identical at $10 per million input tokens and $50 per million output tokens - Anthropic is not charging a premium for the unrestricted configuration; it is restricting who can buy it, not how much it costs.

Fable 5 vs Mythos 5 at a glance

	Claude Fable 5	Claude Mythos 5
Model ID	`claude-fable-5`	`claude-mythos-5`
Availability	Generally available (API, claude.ai, major clouds)	Approved Project Glasswing partners only (~50 organizations)
Safeguards	Full ASL-3/CB-1 stack: cyber, bio/chem, and anti-distillation classifiers	Cyber safeguards lifted; other protections retained
Audience	Developers, businesses, consumers	US government and critical-infrastructure cyberdefense teams
Pricing	$10 / $50 per MTok in/out	$10 / $50 per MTok in/out
Data retention	Standard Claude API policies	Mandatory 30-day retention on all traffic

Data retention and oversight

Mythos-class traffic comes with strings attached. All Mythos 5 usage is subject to mandatory 30-day data retention. Retained data is designated for safety use only, and any human access to it is logged. For organizations doing sensitive vulnerability research, that is a meaningful trade-off - and a deliberate one: Anthropic wants an audit trail on every interaction with its least-restricted model.

Coming next: biology. Anthropic has said a trusted-access program for biology researchers is planned. That configuration would lift the bio/chem safeguards while keeping the cyber classifier in place - a third safety configuration of the same underlying model, tailored to a different vetted audience.

What it means for the industry

TechCrunch framed the launch pointedly: Anthropic released its most capable model "days after warning AI is getting too dangerous." The tension is real, but the structure of the release is the answer Anthropic is offering to it. Rather than a binary choice between shipping a frontier model and shelving it, the Fable/Mythos split treats safety as a configurable layer: one model, multiple safeguard profiles, each matched to a vetted audience and a use case.

That approach has critics. Some Hacker News users have documented false positives from Fable 5's classifiers - a question about mosquito-borne malaria refused, an MRI brain-segmentation task flagged as bioterrorism - and some researchers argue the anti-distillation safeguards are as much about competition as safety. Nathan Lambert of Interconnects called Fable 5 "definitely the smartest model available to the general public," while objecting that "an AI model that gets less intelligent automatically without notifying me is categorically misaligned AI." (Our safety deep dive covers the full classifier system and the criticism.)

Still, the precedent matters. Mythos Preview established that a frontier lab will withhold a model it deems unsafe to release; Mythos 5 establishes the follow-up move - releasing it anyway, but only to the people defending the systems it could otherwise threaten, under logged, time-limited retention. Whether other labs adopt the pattern may shape how the next generation of frontier models reaches the world.

The bottom line

Claude Mythos 5 is not a separate, secret super-model - it is Fable 5 with the cyber guardrails removed, reserved for cyberdefense partners under Project Glasswing. If you are a developer or business user, Fable 5 is Mythos-class capability; the safeguards, not the weights, are what differ. For specs, pricing, and how Fable compares to the rest of the lineup, see our model comparison.