Anthropic's strongest model was completely offline for three weeks. Now it's back. With a big comeback announcement — just not for everyone, and not for free.
Quick recap: what happened?
If you missed June: Anthropic released Claude Fable 5, the first publicly usable model from its new Mythos line. Three days later, the US government ordered Anthropic to shut it down worldwide. The reason: security researchers at Amazon found a prompt that could bypass some of its built-in safeguards — and helped the model hunt for software vulnerabilities. That was enough for the US Commerce Department to impose export controls. Fable 5 disappeared for everyone, free and paid accounts alike.
So is everything fixed now?
Not quite, but better. In late June, the Commerce Department lifted the export controls, and Fable 5 has been running again since — worldwide, with no nationality verification, on claude.ai, in Claude Code, in Claude Cowork, and through the API. Before flipping the switch back on, Anthropic trained a new safety classifier specifically to catch the reported bypass technique. According to Anthropic, it now blocks it in more than 99% of attempts.
Some context: 99% sounds great, but it also means that with enough attempts, something still slips through occasionally. No model is watertight. That was true before the shutdown, and it's still true now.
By the way, Mythos 5 — Fable's quieter sibling model — is back too. Access on AWS, Google Cloud, and Microsoft Foundry is supposed to follow "as soon as possible," though Anthropic hasn't given a firm date for that.
What does it cost me?
Here's where it gets uncomfortable if you're on claude.ai's free tier: Fable 5 isn't available there at all. Full stop.
If you have a paid plan — Pro, Max, Team, or select Enterprise tiers — you can currently use Fable 5 within your normal weekly usage limit, but only up to a fixed share of it. After that, the system automatically switches you to another model. And that window is closing fast: starting July 8, Fable 5 runs exclusively on separately purchased usage credits, billed at API rates — reportedly around $10 per million input tokens and $50 per million output tokens. For comparison, that's dramatically more expensive than the current default model, Sonnet 5, which is free for everyone.
Is it actually worth it for you?
Short and honest: if you're just getting started with AI, Fable 5 is not your entry point. You'd be paying extra for capabilities you probably aren't pushing the limits of yet anyway. Sonnet 5 handles the vast majority of everyday tasks — writing, answering questions, understanding code — extremely well. And it costs nothing.
Fable 5 makes sense for two groups: developers working at the absolute performance ceiling, for whom $10 per million tokens is a rounding error in the project budget. And curious paid subscribers who want to use the short window before July 8 to try it while it's still covered by their plan.
Everyone else: you're not missing out. Go see what your current free model can actually do — probably more than you think.