2 min read AI-generated

Claude Fable 5 and Mythos 5: Anthropic Brings the Mythos Class to Everyone

Copy article as Markdown

Anthropic has released Claude Fable 5 — the first Mythos-class model available to everyone. State of the art on nearly every benchmark, with built-in safeguards. And for cyber defenders, there's Mythos 5 on top.

Featured image for "Claude Fable 5 and Mythos 5: Anthropic Brings the Mythos Class to Everyone"

Ever since the Mythos leak back in March, we’ve known Anthropic has a model that sits a whole tier above Opus. For a long time it was locked away — available only to a handful of cyber defenders in Project Glasswing. As of June 9, that’s over: Anthropic has released Claude Fable 5, the first Mythos-class model that’s open to all.

What Fable 5 can do

Anthropic says Fable 5 is more capable than any model they’ve ever made generally available. It’s state of the art on nearly every tested benchmark: 95 percent on SWE-bench Verified, 80 percent on SWE-bench Pro, top of the pack on Cognition’s FrontierCode. And the pattern is clear — the longer and more complex the task, the bigger the lead.

One example that sticks: Stripe had Fable 5 run a migration across a 50-million-line codebase that would have taken a full team over two months by hand. Fable 5 did it in a day. On vision it’s leading too — it can rebuild a web app’s source code from screenshots alone, and it beat Pokémon FireRed with a vision-only harness, something earlier models needed a whole scaffold of helper tools to manage.

The catch: safeguards

A model this powerful brings risks — especially around cybersecurity and biology. So Anthropic shipped Fable 5 with classifiers: queries about cybersecurity, biology, chemistry, or distillation aren’t answered by Fable but handed off to Opus 4.8. That happens in fewer than five percent of sessions, according to Anthropic. The settings are deliberately conservative — sometimes harmless requests get caught, and that’s meant to improve over time.

For a small group of cyber defenders and infrastructure providers, there’s also Mythos 5 — the same model, but with the safeguards lifted in some areas. Anthropic says it has the strongest cybersecurity capabilities of any model in the world, and it runs initially through Project Glasswing in collaboration with the US government.

Price and availability

Both models cost $10 per million input tokens and $50 per million output tokens — less than half the price of Mythos Preview. On the API, Fable 5 is fully available right away.

My take

This is the biggest leap I’ve seen from Claude — and at the same time the most cautious launch. Anthropic is shipping a model it openly calls potentially dangerous, and building the brakes in from the start. You could write that off as marketing. I find it more honest than the industry norm. If you really want to test Fable 5, reach for the long, multi-step tasks — that’s exactly where the gap to Opus is widest.


Sources: Anthropic, TechCrunch, VentureBeat