Key Points
- Anthropic released Claude Fable 5 and Claude Mythos 5 on June 9, 2026.
- Both share one underlying model; safety classifiers separate the two releases.
- API pricing is $10 per million input tokens and $50 per million output.
- Tops FrontierCode and Hebbia finance benchmarks; matches humans on protein design.
Anthropic released two versions of the same frontier model on June 9, and the difference between them is the whole story. Claude Mythos — the model the company spent the spring teasing — ships in two forms: Claude Fable 5, the general release wrapped in new safety classifiers, and Claude Mythos 5, the same model with those guardrails selectively lifted for vetted researchers. The weights are identical. What you are allowed to do with them is not.
Two names, one model
Fable 5 is the version almost everyone will use. It carries new classifiers covering cybersecurity, biology and chemistry, and distillation attempts — the guardrails that keep a general-purpose model from being turned into a weapon or cloned into a cheaper competitor. Mythos 5 strips some of those constraints for a narrow set of partners: Anthropic's Project Glasswing collaborators and select biomedical researchers in an emerging trusted-access program, who get the model's full reach for autonomous drug design and genomics work. It is capability gating by access rather than by model — a design that treats the safeguards, not the intelligence, as the thing being rationed.
A third, separate name entered the picture in July: Claude Mythos Preview, an unreleased research-only system distinct from Mythos 5, which Anthropic said found new attacks against the HAWK post-quantum signature scheme and a reduced-round version of AES.
What it does
On capability, Anthropic is claiming the top of nearly every benchmark it tested. The model posts the highest score on Cognition's FrontierCode evaluation, leads Hebbia's Finance Benchmark, and, in the lab, designs proteins on par with skilled human operators while out-reasoning specialized models on assembly-prediction tasks. It handles context in the millions of tokens, reconstructs working code from screenshots, and generates novel hypotheses in molecular biology. Stripe, an early tester, said Fable 5 "compressed months of engineering into days" on a codebase migration — the kind of line that reads as marketing until you price out what months of senior engineering actually cost.
The price and the window
The API runs $10 per million input tokens and $50 per million output, available now under the claude-fable-5 model ID. Subscribers on Pro, Max, Team, and Enterprise get Fable 5 included through June 22, after which it draws on usage credits — a soft on-ramp that gets the model into workflows before the meter starts. For a company in active IPO preparations, the move reads as both a product launch and a revenue argument: the most capable model Anthropic has shipped, priced to be used and gated to stay safe, released the same week the market is deciding what the company is worth. The floor under that pricing keeps dropping — five weeks later Moonshot AI shipped Kimi K3 at $3 per million input tokens and $15 per million output, with the weights themselves due out under a Modified MIT license.
Anthropic priced its next release well below this tier: just over six weeks later it launched Claude Opus 5 at $5 and $25 per million tokens, coming close to Fable 5's frontier intelligence at half the cost.
Source: Anthropic
Comments