Anthropic Puts Its Mythos-Class Model in Public Hands as Fable 5

Anthropic has done something it has so far held back from: it put a model from its most powerful class into the hands of ordinary users. On June 9 the company released Claude Fable 5, the first generally available model in the same family as Mythos, the system whose ability to find and exploit software vulnerabilities rattled the cybersecurity world earlier this year. Fable is Mythos with guardrails, and the timing is striking, arriving days after Anthropic itself warned that AI is getting dangerous enough to worry about.

What it can do

The capability claims are large. Anthropic says Fable 5 is state of the art on nearly all tested benchmarks, with standout results in software engineering, knowledge work, vision, and scientific research. On some measures the company reports it scoring more than 10 percent above Claude Opus 4.8, which until now was its strongest public model. In Anthropic's own words, "Fable's capabilities exceed those of any model we've ever made generally available."

That is a vendor describing its own product, so the benchmark talk deserves the usual pinch of salt until independent testers weigh in. What is not in doubt is the distribution. Fable 5 went live the same day on AWS, Microsoft Foundry, and GitHub Copilot, which is about as broad a launch as a frontier model gets.

The safeguards are the point

Here is the part that makes this a release worth pausing on. Anthropic says it could only put a Mythos-class model out in public because of new safety machinery. When Fable 5 is asked something in a genuinely sensitive area, specifically cybersecurity, biology, chemistry, or model distillation, it declines and quietly falls back to Claude Opus 4.8 for that response. The most capable model handles the everyday work; the older, better-understood one handles the queries that could do harm.

Alongside the public Fable, Anthropic also shipped Claude Mythos 5, the same underlying model with some of those safeguards lifted for vetted enterprise and research use. The two-tier setup is the whole strategy: one version for the public with the dangerous edges sanded off, one for trusted hands with more of the raw capability intact.

Getting it, and paying for it

For now Fable 5 is included at no extra cost on Anthropic's Pro, Max, Team, and seat-based Enterprise plans through June 22. After that, access shifts to buying additional compute credits. API pricing lands at $10 per million input tokens and $50 per million output tokens, which Anthropic notes is less than half what the earlier Mythos Preview cost.

The interesting tension sits in plain sight. As TechCrunch noted, this is a company that spent recent weeks cautioning that frontier systems are becoming too powerful to release casually, now releasing its most powerful public model to date. Anthropic's answer is that the safeguards are exactly what make the release responsible. Whether that holds up depends on how well the fallbacks resist people determined to get around them, which is the test every gated model eventually faces.