What Happened To Fable 5

Last week this was the most powerful AI model anyone could use. It sat on top of every leaderboard, ahead of GPT-5.5, ahead of Gemini. Four days later the US government pulled it offline. The model is Claude Fable 5, and how it went from launch to shutdown in four days is stranger than it sounds.

Here's what it actually is, how it compared on the official numbers, why the government stepped in, and which model I'd use right now while this plays out. Every claim links to its source at the bottom so you can check it yourself.

What Fable 5 is

Anthropic launched Claude Fable 5 on June 9th. It's the first model from a new tier they call Mythos class, which Anthropic says sits above the Opus models in raw capability. This wasn't a small bump. It was the most powerful thing they'd ever put in front of the public.

One detail matters for the rest of the story: Fable 5 comes in two versions. Fable 5 is the public one, the strong model with safeguards switched on. Ask it about cybersecurity, biology, or chemistry and the safe version quietly hands that request to the older Opus 4.8. Mythos 5 is the same model without the safeguards, and that one isn't public. It only goes to a small group of cybersecurity partners and a few researchers.

So Anthropic shipped its most powerful model, but the public version had a safety net built in. Hold on to that, because it's the whole reason this happened. Pricing was what you'd expect for a premium tier: $10 per million tokens in, $50 per million out.

The benchmarks

This is where it gets interesting. On SWE-Bench Pro, which gives the model real software engineering tickets to solve, Fable 5 scored 80.3%. For comparison, GPT-5.5 got 58.6, Gemini 3.1 Pro got 54.2, and even Anthropic's own Opus 4.8 was on 69.2. Fable 5 wasn't a bit ahead, it was more than 20 points clear of the next model.

On the harder coding tests the gap got wider. On the difficult problems Fable 5 solved around 29% while GPT-5.5 was under 6. And on Chatbot Arena, where real people vote on which answer is better, it went straight to number one. On the public numbers, for those few days, this was simply the best model you could get. Which is what makes the next part so unusual.

The suspension

Four days after launch, you couldn't use it anymore. Here's the timeline, because a lot of versions are going around.

On Friday, June 12th, the US government called Anthropic around one in the afternoon. A few hours later, at 5:21, a formal letter arrived: an export control directive from the Commerce Department, citing national security. The wording was specific. It told Anthropic to suspend all access to Fable 5 and Mythos 5 for any foreign national, inside or outside the US, including Anthropic's own employees who aren't US citizens.

Read that carefully, because this is where most people get it wrong. The order was about foreign nationals, not literally every user. But to be sure they were complying, Anthropic switched off both models for all customers worldwide. That's why you saw headlines saying the US government banned it for everyone. The shutdown was global, but the order underneath it was about citizenship.

So why did it happen? That's the frustrating part, because the reason is vague. The government believes someone found a way to jailbreak Fable 5 and get around the safeguards. The letter didn't include the details. Anthropic says it only got verbal evidence of what it called a narrow, non-universal jailbreak, and it didn't agree with the decision. It complied, but said publicly that a narrow potential jailbreak shouldn't be reason to recall a model already used by hundreds of millions of people. Every other Anthropic model stayed online. Only these two went dark.

The feud behind it

If this were a one-off it would be odd enough, but it isn't. It's the latest round of a fight that's been going since February.

Back in late February the administration told federal agencies to stop doing business with Anthropic. Then the Pentagon labelled the company a supply chain risk, which basically means the government decided its tech wasn't safe to rely on. Where did it come from? Talks over a defense contract had broken down. Anthropic had two lines it wouldn't cross: it didn't want its models used for mass surveillance of US citizens, and it didn't want them used for autonomous weapons.

So in March, Anthropic sued the Department of Defense. It won an early injunction in one case, then lost an appeal in another, and it's been going back and forth in court for months. The president hasn't been quiet about it either, calling the company, in his own words, "left-wing nut jobs" trying to strong-arm the defense department. So when the June suspension landed, it didn't come out of nowhere. It came on top of months of tension.

Right now both sides are talking. On Monday the 15th, senior Anthropic people including CEO Dario Amodei met the commerce secretary in Washington to try to sort it out. One thing to get right, because you'll hear it loosely online: that's a meeting with administration officials and the Commerce Department, not a confirmed sit-down with the president. Asked about the visit, the president said he didn't know anything about it. As I write this, we don't know how the meeting went, or whether Fable 5 is coming back.

What to use right now

Fable 5 and Mythos 5 are off, for everyone. If your workflow leaned on Fable 5 during those few days, you've already had to move back. The most capable model you can use today is Opus 4.8. For Claude Code, that's the one I'd use until this resolves.

If you want a quick reference for which model to use for what, I made a free cheat sheet: AI model cheat sheet.

One more thing worth watching. The order was specifically about foreign nationals. So even if Anthropic and the government strike a deal, it's plausible Fable 5 comes back to US citizens first, before the rest of the world. If you're outside the US, like I am, keep an eye on that.

The bigger lesson is simple. If you're building on someone else's model, you're really building on that model's relationship with its government. It can be the best model on Monday and gone by Friday for reasons that have nothing to do with the model itself. So keep your setups flexible. Don't hard-wire one model into everything you run.

Sources

If you're building on these models and want a setup that can swap them out when something like this happens, get in touch. See our AI consulting service for how we work.