So Claude Mythos finally arrived. Sort of. It's complicated.
Earlier today, June 9, Anthropic released two new models at the same time: Claude Fable 5 and Claude Mythos 5. Same brain, different leashes. Fable is the version you and I can actually use. Mythos is still locked behind Project Glasswing, available only to a handful of cybersecurity partners and, soon, biology researchers.
If you're wondering why two names for what's essentially the same model, here's the explanation Anthropic gives in a footnote: Fable comes from the Latin fabula, "that which is told," which is roughly the same as the Greek mythos. The model is the same. The story you're allowed to tell with it is what changes.
I find this kind of charming. I also find it slightly worrying. Stay with me.
What Fable 5 actually does
This is the bit where I should be measured and professional. Reader, I cannot. The benchmarks are absurd.
Stripe ran it on a 50 million line Ruby codebase and said it completed a migration in a day that would have taken their team two months by hand. Two months, compressed into a day. The model also beat Pokémon FireRed using only screenshots, with no helper tools, when previous Claude models needed extensive scaffolding just to get through the game at all. It built a working solar system simulation, derived the orbital physics from first principles, and used the result to predict solar eclipses. Just for fun, presumably.
For marketers, the bits that matter most are this. It's now genuinely capable of senior-grade analytical work, the kind of thing you'd hire a smart analyst to do. Vision is finally a real production tool, so you can hand it a screenshot of a competitor's webpage and have it rebuilt as code, which is a quietly enormous shift for design audits and competitor research. And it now stays focused across millions of tokens with persistent memory between sessions, which means full campaign reviews, big content audits, and multi-document research are properly on the table.
What you and I should do about it
For the next two weeks, until June 22, Fable 5 is included on Pro, Max, Team, and Enterprise plans at no extra cost. After that, it shifts to usage credits until Anthropic builds enough capacity to roll it back into the standard plans.
Translation: there is a free window. If you've been curious, the next fortnight is when to play. Run your hardest problem, your weirdest brief, your boring monthly report through it and see what happens. The free clock is ticking.
The Mythos bit, for context
Mythos 5 is the same model with the cybersecurity safeguards switched off. It's restricted to vetted partners through Glasswing, which Anthropic runs with the US government. The reason given is essentially: this thing is good enough at finding software vulnerabilities that putting it in everyone's hands would be irresponsible.
Read that again. The company that built the model has decided, voluntarily, that its full power is too dangerous for a public release. That's not nothing. We've spent two years hearing every AI lab promise their next model will be the best ever. This is the first time I've seen one of them say "and also, we've decided you can't have most of it."
There's a marketing lesson buried in here too. Restraint, when paired with a genuinely premium product, is a hell of a positioning move. Imagine the energy of a brand saying "we made something so good we won't sell it to most of you." Anthropic gets to look responsible AND aspirational at the same time. That's a rare double.
So where does this leave us?
For the next fortnight, the best AI model available to the general public is free on the plan most of you are already paying for. After that, it gets metered. Either way, the gap between what was possible last week and what's possible today just widened by a noticeable amount.
If you've been waiting for "the moment" to start taking AI seriously in your workflow, this is a reasonable candidate.
We're already running ours through it. I'll let you know what breaks.