Claude Fable 5 and Mythos 5: The Most Powerful AI Anthropic Built

Quick Summary: Claude Fable 5 is Anthropic’s most powerful publicly released AI model, launched June 9, 2026, in a new top tier called “Mythos-class.” Claude Mythos 5 is the same model with safeguards removed for vetted partners. Both topped major benchmarks, but a US government directive suspended access on June 12, with no confirmed return date. Use Claude Opus 4.8 meanwhile.

Claude Fable 5 was so good that the US government stepped in and told Anthropic to turn it off.

That actually happened within 72 hours of launch. You may already see there is a tiny notification upside your Claude’s prompt bar “Claude Fable 5 is currently unavailable.”

Let me walk you through what is Claude Fable 5, why people couldn’t stop talking about them, how a flagship launch turned into one of the strangest weeks the AI industry has had in a while.

What is Claude Fable 5?

Claude Fable 5 is Anthropic’s most powerful model ever made available to the general public.

It launched on June 9, 2026, and it belongs to a brand-new tier Anthropic calls “Mythos-class”- a step above the Opus models that sat at the top of the lineup until now.

In plain terms? It’s the smart one.

Think of Anthropic’s models like a team.

Haiku is the fast junior who knocks out quick jobs.

Sonnet is the dependable mid-weight.

Opus has been the senior expert.

Fable 5 is the new hire who walks in on day one and quietly outperforms everyone on nearly every test they throw at it.

Anthropic says Fable 5 is state-of-the-art on almost all the AI capability benchmarks it was measured against, with standout results in software engineering, knowledge work, vision, and scientific research. And here’s the interesting part: the harder and longer the task, the bigger its lead gets.

Most models start wobbling when you give them something sprawling and multi-step. Fable 5 seems to do the opposite: it stretches its legs.

How is Claude Mythos 5 Different from Fable 5?

This trips people up, so let me make it simple. Fable 5 and Mythos 5 are the same underlying model. Same brain with same raw ability.

The difference is the seatbelt.

Fable 5 comes with built-in safety filters. Think of them as guardrails that stop the model from helping with dangerous stuff (I’ll explain how those work a bit later). Mythos 5 is the same model, but with some of those guardrails switched off.

Because of that, Mythos 5 isn’t something you or I can sign up for. It’s handed to a small, carefully checked group, mainly cybersecurity experts and the people who protect critical systems like power grids and networks. They get access through an Anthropic program called “Project Glasswing,” which works closely with the US government.

Mythos 5 is said to be the best in the world at cybersecurity tasks, and that’s the whole reason it’s kept on a tight leash. In the wrong hands, those same skills could do real damage.

The names are a nice little touch, too. “Fable” and “mythos” both basically mean “a story that gets told.” Same idea, two different versions of one model. The only thing that separates them is the safety layer.

Use Odysseus: No Cloud Storage Attached and No Tracking

Claude Fable 5 Benchmarks: Why the Results Stunned Everyone?

Because the early numbers were genuinely a bit silly. In a good way. Let me give you the highlights without drowning you in charts.

1. Coding That Compresses Months into Days

Stripe got early access and reported that Fable 5 took a job that would’ve eaten one of their engineering teams over two months: a code migration across a 50-million-line Ruby codebase. And finished it in a single day!

It’s also more efficient with its words (tokens), which matters because tokens cost money. On Cognition’s frontier coding evaluation, Fable 5 came out on top among leading models even when running at medium effort. So, it’s not just smart; it’s economical with it.

2. Knowledge Work, Finance, and the Boring Stuff That Actually Pays

This is the part marketers and operators should care about.

On a senior-level finance reasoning benchmark from Hebbia, Fable 5 posted the highest score of any model, with big jumps in reading documents, interpreting charts and tables, and actual problem-solving. A trading firm called IMC said it basically aced their analysis evaluations across the board.

If your work involves messy spreadsheets, dense reports, or “read these 40 PDFs and tell me what matters,” this is the headline you wanted.

3. Vision, or “The Model That Beat Pokémon by Just Looking at The Screen”

This is my favourite story from the launch.

To play the game Pokémon FireRed, older Claude models needed a lot of help: extra tools feeding them maps and hints about what was happening on screen.

Fable 5 didn’t need any of that. It finished the game just by looking at the pictures on the screen, the same way you or I would.

And it’s not only good at games. Fable 5 can do genuinely useful visual work, too, like reading exact numbers off a complicated chart or looking at a website screenshot and writing the code to rebuild it.

The fun trick and the great skill come from the same ability: it can actually understand what it’s looking at.

4. Science That Surprised The Scientists

Anthropic’s drug-design experts used Mythos 5 to speed up parts of their work by about 10 times. The model handled jobs that normally need a trained scientist to do.

It also came up with its own ideas in biology. When scientists compared the model’s ideas against human ones without knowing which was which, they liked the model’s ideas about 80% of the time.

One of those ideas was even later proven right by a separate lab working on the same problem. That’s a big deal. It means the model wasn’t just guessing or repeating things; it was suggesting real ideas that actually turned out to be correct.

What about The Safeguards Everyone Keeps Mentioning?

Right, the safety filters. This is where Fable 5 gets interesting.

Anthropic gave Fable 5 a separate set of AI “checkers” that watch for risky requests, mostly around hacking, biology and chemistry, and people trying to copy the model. When one of these checkers spots something risky, Fable 5 doesn’t flat-out refuse. Instead, it quietly passes the job to a slightly less powerful model (Opus 4.8) and tells you it did so.

Anthropic admitted they set these filters to be extra careful on purpose, which means they sometimes block harmless questions by mistake. The company says this happens in fewer than 5% of chats, so more than 95% of the time, you’re getting the full-strength Fable. Still, plenty of users complained the filters felt too strict, and Anthropic agreed that was fair and promised to loosen them over time.

It’s a real trade-off: release a powerful model quickly, or spend longer making the filters perfect. They chose quick and careful. You can disagree with the choice, but at least they were upfront about it.

The Plot Twist: the US Government Pulled the Plug

Now the part that turned a product launch into a news story.

On June 12, just three days after launch, the US government stepped in. They said it was about national security, and they told Anthropic to shut off Fable 5 and Mythos 5 for everyone. Anthropic had to switch both models off straight away. (All the other Claude models kept working fine, so this wasn’t Claude going down completely, just these two.)

Why did this happen? The government thought someone had found a trick to get past Fable 5’s safety filters.

Anthropic looked at the so-called trick and said it was pretty basic: it just asked the model to read some computer code and point out the flaws in it. And the flaws it found were small ones that people already knew about, the kind of thing other AI models (like OpenAI’s GPT-5.5) can already do anyway.

So now Anthropic is obeying the order, but it doesn’t agree with it. Their point is simple: shutting down a model that hundreds of millions of people use, over a small problem that didn’t actually hurt anyone, is a huge overreaction. And if that became the normal way of doing things, no AI company would ever be able to launch anything new.

This is the real tug-of-war: incredibly powerful AI on one side, and governments worried about safety on the other. One popular online post summed up the week perfectly — it felt less like a new product coming out and more like a major world event. And honestly? Hard to disagree.

How Much Does Claude Fable 5 Cost, and Where Can You Use It?

When it was live, both Fable 5 and Mythos 5 cost $10 per million words of input and $50 per million words of output (that’s roughly what “tokens” means), less than half the price of the older Mythos Preview. For a model this powerful, that’s cheap, and it showed Anthropic wanted to get this into as many hands as possible.

A few details for the builders:

• The model’s name in the API is claude-fable-5 (and claude-mythos-5 for the restricted one).
• It can handle a huge amount of text at once, up to a million words in one go.
• It was available through the Claude API, AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry.
• Because the safety filters can block a request, Anthropic built in a backup system so developers can automatically retry on another model when that happens.

So, When Can You Actually Use Claude Fable 5?

Let me give you the straight answer, because this is what everyone really wants to know: right now, you can’t and there’s no confirmed date for when it comes back.

As of mid-June 2026, both Fable 5 and Mythos 5 are switched off worldwide. Anthropic’s status page just says they’re “monitoring” the situation, and their only promise is that they’re working to bring it back “as soon as possible.” No date. Anyone giving you a hard return date right now is guessing.

Here’s a realistic read on the timing:

• Best case, a few days. If Anthropic quickly works out a fix with the government (like limiting access to verified US users), it could come back fast. The company clearly thinks this is a misunderstanding and wants it live again.
• Slower case, a few weeks. If the legal back-and-forth drags on, it could stay off for a while.

Two things worth keeping in mind:

1. There was a separate plan for paid subscribers: Fable 5 was meant to be free on Pro, Max, Team, and Enterprise plans through June 22, then switch to paid credits after. That schedule is on hold too, so don’t be surprised if the dates change once it’s back.
2. The fastest way to know the moment it returns is Anthropic’s official status page and newsroom — not the rumour mill.

What Does Claude Fable 5 Actually Mean for You?

If you’re a builder, a marketer, or just trying to keep up, here’s the takeaway without the drama.

The bar just jumped. Even with access paused, Fable 5 showed what’s coming next: AI that can handle long, complex jobs on its own that older models simply couldn’t finish. If your work involves big projects, deep research, or heavy document analysis, that’s where things are heading, fast.

The “safety vs speed” debate is now a real business issue, not just talk. A government just showed it can pull a paid AI model overnight.

So if you’re planning to build your business on one model, “what happens if it disappears for a week?” is no longer a silly question. Always have a backup option ready; Anthropic literally built one in for this exact reason.

And the falling price is good news for you. Top-tier power at less than half the old cost tells you where this is going. The smart move isn’t to chase whatever model is best this week, it’s to build things so you can switch models easily, because the leader will keep changing.

FAQs on Claude Fable 5

The Bottom Line

Claude Fable 5 had one of the most dramatic debuts in AI history. A model that truly raised the bar, priced for everyone, then switched off by a government order before most people even got to try it properly.

Whatever happens next, this launch taught us two things worth remembering. First, AI is moving faster than the rules around it can keep up. Second, the days of “just pick the best model and build on it forever” are over; the ground moves now, sometimes overnight.

So keep an eye on this one. If Anthropic is right that it’s all a misunderstanding, Fable 5 will be back, and you’ll want to be ready to put it to work. And if you’re building anything serious on these tools, take the hint from this past week: never bet your whole business on a single model staying exactly where you left it.

This article reflects events as of mid-June 2026 and will be updated as the situation develops.