Anthropic’s June 9 release of Claude Fable 5 introduced something more interesting than another benchmark leader: a two-track distribution model. Fable 5, the public version, is what Anthropic calls a Mythos-class model made safe for general availability. Claude Mythos 5 — the same underlying model with safeguards lifted in specific areas — is restricted to approved organizations: cyberdefense partners in a government collaboration called Project Glasswing and select biology researchers.
Five weeks in, the shape of this release is worth understanding, because it is likely the template for how frontier labs ship their strongest models from now on.
The facts
Fable 5 sits above Opus 4.8 in Anthropic’s lineup and is described by the company as its most intelligent generally available model. API pricing is $10 per million input tokens and $50 per million output — double Opus 4.8’s $5/$25. It is available through the API, the Claude apps, and Amazon Bedrock.
The subscription story had a twist. From launch through June 22, Fable 5 was included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost — a two-week taste. On June 23 it came off those plans, and using it now requires usage credits on top of the subscription. If your workflow got attached to Fable 5 during the free window, you have already discovered this the expensive way.
Mythos 5 is not something you can buy. Access goes through Anthropic’s approval, and coverage from VentureBeat frames the split plainly: the public gets the capability with guardrails; a vetted group gets it with fewer, for defensive security and research work where the guardrails themselves are the obstacle.
Why the two-track structure matters
For years the implicit deal was that everyone got the same model, and safety was a property baked into it uniformly. The Fable/Mythos split makes a different deal explicit: capability and permission are now separate axes. The lab decides not just what the model can do, but who is allowed to use which parts of what it can do.
You can read that two ways, and both are probably right. The charitable reading: dual-use capability in areas like cybersecurity genuinely is different in the hands of a hospital’s defense team than in the hands of an anonymous API key, and tiered access is the honest way to serve one without arming the other. The wary reading: access tiers create a second scarcity — beyond price — and labs now have a governance lever whose criteria are theirs alone to set. Who counts as “approved” is a business decision wearing a safety coat, and there is no appeals process.
Either way, competitors are watching. If the two-track structure holds up commercially and politically, expect it to become the standard shape of frontier releases — a public tier, and a documented-identity tier above it.
What it means in practice
For most builders, three things. First, the practical ceiling of “the best model you can just buy” is Fable 5 at $10/$50, and at that price it is a specialist tool: reserve it for the requests where judgment failures are expensive, and let cheaper models carry volume. Second, the June 23 removal from subscriptions is a reminder that launch-window inclusion is marketing, not a commitment — never architect a workflow around plan inclusion you have not seen survive a billing cycle. Third, if you work in security research or biosecurity and the public model’s refusals block legitimate work, a formal access path now exists where previously there was none; the announcement page describes the program.
Facts and pricing in this piece were checked against Anthropic’s announcement and the coverage linked above as of 2026-07-17. Prices and plan inclusions in this market change quickly — verify against the official pages before making decisions that depend on them.