Will Fable 5 and Mythos Ever Come Back?

Someone has registered isfable5back.com. The whole site is one word, “No,” with a little timer counting how long the best coding model anyone’s used has been switched off. A week now. I check it more than I’d like to admit.

If you missed it, the US Commerce Department ordered Anthropic to cut off Fable 5 and Mythos 5 for any foreign national on the 12th, at 5:21pm Eastern, citing national security. There’s no clean way to tell a foreign national from a US person across a few hundred million accounts on a Friday afternoon, so Anthropic did the only thing it could and pulled both models for everyone. I wrote about the politics of that already, the part where Dario spent months asking for exactly this power and then got hit with it. This post is the more selfish question. The one every developer who’d built Fable 5 into their day is actually asking. Is it coming back, and when it does, will it still be the thing we lost?

Whether it comes back is the easy bit to call, and I’d bet the house it does. It’s an export-control order over a jailbreak that spooked the government, not a recall for anything Anthropic did wrong, so there’s an actual way out of this. Anthropic’s whole public posture is that the order is a misunderstanding and they’re “working to restore access as soon as possible.” The administration’s line, as far as anyone can piece it together, is that Anthropic can fix the jailbreak or de-deploy, and once the fix lands the order gets lifted. So there’s a defined path back. That’s very different from a model being killed off for good. The New Stack quoted someone close to it saying the ball is in Anthropic’s court, which is corporate-speak for “they know how to make this end.”

What nobody will commit to is a date. No timeline, no “back by Friday,” nothing. Refunds are already going out for a product that vanished four days after launch, which tells you Anthropic isn’t expecting this to resolve in hours. My honest guess is weeks, not days and not months. They’ve got an IPO breathing down their neck and every extra day Fable 5 stays dark is a day the prospectus looks worse. The incentive to get this sorted is enormous.

What I actually worry about is whether the model that comes back is the same one that left.

A lot of people are bracing for disappointment here, and they might be right. The fix the government wants is the closing of a jailbreak. The way you close a jailbreak is more guardrails, more refusals, more of the model second-guessing what you asked it to do. So even though nobody is “nerfing” Fable 5 in the sense of shrinking it or swapping in a weaker version, the realistic outcome is a model with the safety dial turned up and a few more “I can’t help with that” walls than the one that launched. Restricted access for some users is on the table too.

That matters because the jailbreak in question was the model reading a codebase and pointing out security flaws. That’s not some exotic dark-arts capability. That’s the exact thing half of us were using it for. If the fix is teaching Fable 5 to be more nervous about reading code and flagging bugs, it lands right on the use case that made the model worth having. I don’t think it’ll be lobotomised, but the version that comes back will probably flinch at a few things the launch one handled without blinking, and we’ll all spend a week working out which prompts it suddenly hates.

Then there’s the question I keep seeing people dodge. Was Fable 5 actually that good, or are we just mourning the hype?

It was that good. I went back through the numbers because I didn’t trust my own enthusiasm, and they hold up. On SWE-Bench Pro it scored 80.3%, with Opus 4.8 sitting at 69.2%, GPT-5.5 at 58.6%, and Gemini 3.1 Pro at 54.2%. That’s a different tier, not a rounding error. Every ran it through their Senior Engineer benchmark, the hardest coding test they’ve got, and Fable 5 scored 91 out of 100, into the range of actual human engineers. Opus 4.8 got 63 on the same test. GPT-5.5 got 62. Stripe said it pushed a migration across a 50-million-line Ruby codebase in a single day, work they’d budget a team more than two months for.

I felt that gap before I saw the numbers. Going from Opus 4.8 to Fable 5 was the biggest single step up I’ve felt from one of these things in a long time. It changed what the work felt like, handing a task off instead of standing over it. So no, the grief isn’t hype. People are cut up because it was the best tool any of us had, and it got switched off by a letter, not by a better competitor.

I’ll put one asterisk on the love-in though, because pretending it was flawless would be its own kind of hype. On the Agent Security League leaderboard, Fable 5 with Claude Code landed mid-table. 59.8% on functional passes and a fairly grim 19.0% on the security ones. Read that next to why it got pulled and there’s a dark joke in there somewhere. The model is a monster at finding flaws in your code and distinctly average at not introducing them in its own. Brilliant and a bit reckless, which is about the most human thing about it.

So where does that leave those of us refreshing a website that just says “No”? It’s coming back. It probably comes back a touch more cautious than it left. And for the few days we had it, it lived up to the noise. If you want the politics of how absurd it is that we’re here at all, that’s the other post. This one’s simpler. I miss the thing, the thing was good, and I’d like it back before the timer on that website needs a second digit.

Sources: