Claude Fable

Well. Damn.

Last week Friday my livestream was about the rather wonderful Claude Fable 5.

Which then pretty much immediately got pulled/banned/nuked from orbit.

Leaving me in a bit of a predicament - today’s newsletter is based on that last livestream. But it’s about a now banned model. Bugger.

Here’s the plan:

• you get this newsletter (obviously, you are reading it now!) about Fable to fill in the “what is it?”

• today (Monday) I’ll do another livestream on its removal/banning.

The situation may change again by then. Or even before you get this newsletter (I’m writing it Sunday and it comes out Monday)! But we’ll do the best we can considering!

If anything these can be some musings on what a model of this power means. How we change how we work with it. What this means for work in general. What this means for access to AI.

These are all themes that are very current. And even if we don’t have access to Fable right now we will have similar powered models as standard in the next year.

What is Fable?

Fable 5 is the new model from Anthropic. It’s a cut down version of Mythos, which you may have heard of. Which makes this a little confusing.

Mythos 5 is the high-risk, full-fat version. Anthropic says it is restricted to Project Glasswing partners, cyberdefenders, infrastructure providers, and soon some biology researchers.

Fable 5 is the public version. Same Mythos-class capability, more guardrails. So it’s a neutered version that (theoretically) cannot be used for evil.

The Claude docs say Fable is built for demanding reasoning and long-horizon agentic work. And they aren’t lying - it’s good. Very very good.

It’s also eye wateringly expensive: $10 per million input tokens, $50 per million output tokens. And only available on the subscription plan until the 22nd June (update: this is up in the air because of the current ban!).

We knew it would be good. That’s not terribly surprising. New model = pretty great is no longer news in the world of AI.

The more interesting part is what the model does to the human job.

Because this does not feel like a slightly better chatbot. It feels like the point where the work moves from "prompt the model" to "commission the work".

The Model You Don't Prompt

This is a transition we’ve been seeing since the beginning of 2026. We’re moving away from prompting and more to commissioning.

A prompt is "write me a blog post about X” and giving it a bunch of specifications.

That’s how we’ve been using AI for the last couple of years.

A commission is "here is the mission, here are the constraints, here is what success looks like, go away for hours and come back with the thing.

That last one is where Fable starts to feel different.

You are not sat there nudging it every three minutes. You are not playing prompt tennis. You are briefing something, walking away, and then inspecting the result later.

That is a very different working relationship.

I saw V talking about writing a Fable wishlist like a kid writing to Santa.

This is what I was using Fable for as well. I dug out my ToDo lists, full of old items that I just haven’t been able to get to. And fed it wholesale into Fable to work out, prioritise and get to work on. And it worked like a charm.

Wizard to Warlock

Let me nerd out for a moment..

In the world of Dungeons and Dragons or any fantasy RPG setting there are various magic users. Wizards, Warlocks, Sorcerers, Mages, Clerics etc. etc.

Wizards and Warlocks are the two I want to focus on here as they are illustrative to how we use AI.

Wizards study magic. They go to magic school and read ancient tomes to hone their craft. They learn special techniques and secrets. They covet and protect knowledge. Wizards are basically magic nerds.

Up until now we AI users have been like Wizards. We study occult techniques, use special incantations, build arcane systems of knowledge. All to coax the best results out of our AIs.

Warlocks instead take a shortcut. None of that boring book learning - no no. They instead cut a deal with a patron - often a dark demonic or fey entity.

Their power comes from a higher entity which they barely control. In fact the entity can entirely forsake them if they displease it…

Using Fable is like contracting with a patron.

We aren’t in control. Fable receives our request then acts in a black box, doesn’t give us much of its reasoning and spits out a scarily good end result.

We aren’t prompting. We’re commissioning it. And we get what it thinks we deserve!

This means… do not use Fable to write an email.

Please.

That is like sending a freight train to pick up groceries.

I saw a lot of people using Fable for mundane easy problems and then not understanding why it’s a big deal. Well, yeah, of course not.

Use it for big, chunky, high-value problems. The things where the cost and slowness are worth it. Business audits. Product strategy. Codebase migrations. Research maps. Security reviews. "Go through everything in this repo and tell me what I am missing." "Help me design a free resource for young people trying to earn in an AI economy." "Audit my business and tell me where I am being an idiot."

Those are Fable-shaped jobs. BIG tasks. BIG problems.

I actually threw a really big one at it - basically saying “Hey Fable youth unemployment is a real problem in the UK and elsewhere and AI is not going to help that going forward. I want to leverage AI with Kyle and AI education to help rather than hinder those 16-25 year olds and help them not lose a decade of their lives”.

The results were phenomenal. Now it’s up to me to act on it.

Increased inequality

The AI now has an hourly wage.

Fable/Mythos (and any equivalent models) are not cheap to run.

They will not be on our subscriptions moving forward because right now our $20/$100/$200 a month subscriptions are causing Anthropic and OpenAI to haemorrhage money.

They’ll instead be on (I suspect) $1000/$2000 a month subscriptions and the API “pay per use” billing.

Someone worked out that Fable/Mythos run costs roughly $40-ish for an hour of work.

That’s more than most humans make. The AI has a higher minimum hourly rate.

And brutally… it’s probably worth it for a business if the output is good enough.

That’s an annual equivalent salary of $80,000 or so (8 hours a day, 5 days a week, 50 weeks a year - although obviously it can work more!).

But the amount you could get done with a model like Fable in that time? Obscenely high. Probably (sorry) much more than an $80,000p.a. human.

Because from the employer's side, the maths gets grim fast. No breaks. No HR. No sick leave. No awkward 1:1s. No "can we circle back next week?" nonsense. Just a model doing focused work for as long as you can afford to run it.

I’m not a huge fan of that future. But I can also see exactly why companies will use it.

You can be morally horrified and commercially forced at the same time. If your competitor can hand a high-value research task to Fable and get a useful answer in three hours, you do not get many points for refusing on principle and spending 3 weeks instead. They’ll eat your lunch.

At the same time this is bad news bears for individuals who (like many) won’t be able to spend this amount on AI. How will individuals compete? How will anything be left? Even crumbs of business and work can be hoovered up by larger business using these AI tools - whereas previously it wasn’t worth their time and effort.

I’m not one to overhype or fear monger. Any one who has read or watched enough of this stuff knows this.

BUT, and I hate to be saying this, we individuals have a narrow window to carve out something for ourselves.

If you are reading this. If you are watching my videos. If you are engaging with this technology you are ahead of the world. That gives you an opportunity over the next couple of years to set yourself up. But that window is not staying open for long - models like Mythos/Fable are a warning shot.

To the Task,

Kyle

Why it matters: : Claude Fable Wizards vs.