Claude Mythos 5 and Fable 5 Explained: What Anthropic Actually Shipped

A new tier above Opus

Anthropic just shipped two models, and the naming is the first thing to untangle. Claude Fable 5 is available everywhere today, while Claude Mythos 5 is restricted to Glasswing partners only. Both belong to what Anthropic calls a Mythos-class tier, a band of models that sit directly above Opus in raw capability. The first member of that tier, Claude Mythos preview, shipped in April through Project Glasswing. Fable 5 and Mythos 5 follow it now.

The most useful clarification: Fable 5 and Mythos 5 are the same model under the hood. The only difference is that Mythos 5 has its cybersecurity safeguards lifted, which is exactly why it does not go out to the general public. Fable 5 keeps those guardrails on so it can be made safe for general use.

Pricing and the free window

Both models cost 10 dollars per million input tokens and 50 dollars per million output tokens. That is double the price of Opus, and plenty of people already find Opus expensive. For context, the original Mythos preview ran five times the cost of Opus, so this generally available version is the more affordable, capped variant.

The timing matters more than the sticker price. From launch through June 22, Fable 5 is included at no extra cost on Pro, Max, Team, and seat-based enterprise plans. On June 23, Anthropic removes it from those plans and moves it to token-based usage credits.

If you want to feel what a tier-above-Opus model does without paying per token, the window before June 22 is the moment to do it.

Anthropic says it intends to restore Fable 5 as a standard part of subscriptions once capacity allows, but gives no firm date.

What the benchmarks claim

Anthropic positions Fable 5 as state-of-the-art on nearly all tested capability benchmarks. The reported gains cluster in a few areas:

• Knowledge work, which took one of the biggest jumps
• Agentic coding, with a noticeable lift on frontier code tasks
• Vision, which strengthens verification loops for editing, slides, and websites
• Legal reasoning, multi-disciplinary reasoning, and biology

- Knowledge work, which took one of the biggest jumps - Agentic coding, with a noticeable lift on frontier code tasks - Vision, which strengthens verification loops for editing, slides, and websites - Legal reasoning, multi-disciplinary reasoning, and biology

Benchmarks always deserve a grain of salt, but the effort-level story is concrete and practical. Fable on low is expected to roughly match Opus 4.8 on extra-high effort while costing less, and max effort gets a modest bump above extra-high for the hardest work.

Mythos, cybersecurity, and caution

Mythos 5 carries the strongest cybersecurity capability of any model, which cuts both ways. A model that good at coding is also good at breaking code, so Anthropic ships it only through Project Glasswing in collaboration with the US government and a narrow set of cyber defenders and infrastructure providers.

There is one more thing worth flagging. The labs now push the idea that you should run loops of agents rather than prompt directly. That can be powerful, but the companies saying it also benefit from heavier token use. Most general knowledge work does not need always-on loops, and firing off open-ended runs is a fast way to hit your session limit before the work is even done.