On 9 June, Anthropic announced Claude Fable 5 and Claude Mythos 5. The interesting part is not that there are two new models. Underneath, there is really only one.
Same model, different doors
Fable 5 and Mythos 5 share the same underlying frontier model. The distinction is in what is switched on.
Fable 5 is that model with safeguards applied so it is suitable for general use. This is what most developers and businesses will access.
Mythos 5 is the same model with those safeguards lifted in specific, tightly controlled domains. Access is restricted to authorised users: Anthropic's "Project Glasswing" partners on the cybersecurity side, and a small set of biomedical researchers on the biology side.
In other words: most of us get Fable 5. A trusted few get Mythos 5. It is a deliberate design: put the frontier capability behind a door, and only open it where the risk can be managed.
What it can do
On ordinary software and knowledge work, the capabilities Anthropic reports are substantial:
| Area | Reported benchmark or claim |
|---|---|
| Software engineering | Tops Cognition's FrontierCode evaluation; Stripe reported compressing months of work into a day on a 50-million-line Ruby codebase |
| Coding tools | Cursor CEO Michael Truell describes it as "the state of the art model on CursorBench," opening "a class of long-horizon problems that were out of reach for earlier models" |
| Finance and knowledge work | Top performance on Hebbia's finance benchmark for senior-level reasoning |
| Vision | Can rebuild a web app's source code from screenshots; state-of-the-art on visual tasks |
| File-based memory | In a benchmark game (Slay the Spire), access to persistent file-based memory improved Fable 5's performance three times more than the same approach improved Opus 4.8 |
These are Anthropic's reported figures, drawn from their announcement.
Where Mythos earns its name
The restricted side is about the life sciences. Anthropic reports protein-design experts accelerating parts of drug design by around 10x, novel molecular-biology hypotheses preferred by scientists about 80% of the time over Opus-class models, and an autonomous genomics run that identified cells across 138 animal species. That is the capability worth gating, and the reason Mythos 5 is not simply on sale.
The safeguards
Fable 5 ships with three new classifier areas: cybersecurity (refused harmful cyberattack planning in external testing), biology and chemistry (falls back to Claude Opus 4.8 on most such requests), and distillation (blocking attempts to extract its capabilities to train rival models). According to Anthropic, the safeguards trigger in under 5% of sessions, with Opus 4.8 handling the response when they do.
Pricing
At $10 per million input tokens and $50 per million output tokens, Fable 5 sits at the top of Anthropic's pricing tier, as you would expect for a frontier release.
- Fable 5 and Mythos 5 are the same underlying model: Fable 5 has safeguards on; Mythos 5 has them lifted in specific domains.
- Mythos 5 is restricted to authorised partners (Project Glasswing and select biomedical researchers). Most developers use Fable 5.
- The benchmark claims are Anthropic's reported figures: treat them as a useful signal, not a guarantee in your specific workload.
- Frontier capability is now genuinely useful for long-horizon work. The headline feature of this launch is as much about governance as raw performance.
Our take
Two things stand out for anyone building on this. First, frontier capability is now useful for real, long-horizon work: not just demos. Second, and more telling, the real story of this launch is governance: who gets which capabilities and how tightly access is controlled. That question is going to matter more, not less, as these models get more powerful.
For most businesses the practical takeaway is simple: the capability ceiling just went up again, and the gap between "a clever demo" and "something that does real work" keeps narrowing. We will be putting Fable 5 through its paces on the long-horizon tasks our clients actually care about and will report back where it genuinely earns its place.
