For two months, Anthropic’s Mythos model existed mostly as a rumor with an access list. Now the doors are opening, at least partway.
Anthropic has released Claude Fable 5 — the first publicly available version of its Mythos architecture. The company first announced Mythos back in April 2026 but kept it locked down over cybersecurity concerns, handing access only to a small group of trusted partners. Last week that circle expanded to hundreds of organizations across 15 countries. This week, a broader version went live for everyone else.
What Fable 5 actually does better
Anthropic says Fable 5 outperforms its previous models in coding, complex analytics, and handling large volumes of information. The third-party numbers back that up. Analytics platform Hex said Fable 5 was the first model to break 90% on its internal benchmark for multi-step analytical tasks. App-building platform Base44 noted its ability to generate complete, functional applications from a single prompt. Genspark reported it outperforms competitors on UI development and game creation tasks.
That’s not marketing copy — those are specific claims from companies that pay to use the thing and have every reason to be honest about whether it works.
The safety trade-off built into the model
Here’s where it gets interesting. Anthropic didn’t just release a more powerful model — it released a more powerful model with a deliberate ceiling built in. Fable 5 will decline certain requests in cybersecurity, biology, chemistry, and model distillation. When it hits one of those walls, it automatically redirects the task to Claude Opus 4.8 instead. You get a response, just not from the most capable version.
Before launch, external researchers spent over 1,000 hours trying to jailbreak the model and found no universal method to bypass its protections. Whether that holds once a few million more people start poking at it is a different question, but it’s a more rigorous pre-launch bar than most AI releases clear.
The privacy policy change nobody asked for
There’s one detail in the fine print that enterprise customers won’t love. Anthropic is introducing mandatory 30-day request log retention for all Fable 5 and Mythos 5 users — including customers who previously had access to a no-logging mode. Anthropic says this is necessary to track new attack patterns and jailbreak attempts.
That’s a reasonable security argument. It’s also a meaningful change for anyone who chose Anthropic specifically because of the stricter data handling. The company is essentially saying: the more capable the model, the more oversight we need to keep around it. You can agree with that logic and still wish they’d asked first.
Who gets it and what it costs
Fable 5 is available via the Anthropic API and Enterprise plans on a pay-per-use basis. Until June 22, it’s also free for Pro, Max, Team, and Enterprise subscribers. After that, Anthropic is switching to a credits system. Pricing sits at $10 per million input tokens and $50 per million output tokens — exactly double the cost of Claude Opus 4.8.
The full Mythos 5 — the unrestricted version — remains available only to organizations that already hold special clearance from Anthropic. That tier isn’t changing anytime soon.
For most developers and businesses, Fable 5 is the most capable AI Anthropic has ever made publicly available. The guardrails are part of the deal — and given what the full Mythos model is apparently capable of, that’s probably the right call.