Claude Fable – Anthropic is rolling out “Fable 5,” a safer version of its high-risk Mythos model class. It comes with strict safeguards, but it’s also token-hungry: users can burn through daily limits in minutes, and access is limited to Pro, Max, and Enterprise customers on
On a Tuesday night, it doesn’t feel like “a new model release” so much as a countdown. Anthropic has begun rolling out its new AI model, “Fable,” which belongs to the same underlying model class as “Mythos,” the company’s most powerful AI model.
The catch isn’t just that Fable is “safer.” It’s also that it’s expensive to run, and Anthropic’s plans for who can use it—and for how long—are deliberately tight.
Anthropic previously said it developed Mythos and that it poses security risks to companies around the world. The company warned at the time that Mythos was powerful enough that it could potentially help bad actors attack public and private software.
In April. Anthropic put it bluntly: “The advantage will belong to the side that can get the most out of these tools.” It added. “In the short term. this could be attackers. if frontier labs aren’t careful about how they release these models. In the long term. we expect it will be defenders who will more efficiently direct resources and use these models to fix bugs before new code ever ships.”.
Anthropic also explained how the same capability could be abused to find and exploit vulnerabilities in apps like Firefox. Because of those risks, it decided to limit access to models like Mythos, offering them only to cybersecurity experts and trusted companies.
Now the company says it has developed guardrails for the same model class—guardrails designed to stop sensitive queries from being exploited. The result is a safer version called “Fable 5.”
Anthropic says Fable 5 blocks or diverts sensitive queries, including those involving offensive cybersecurity, biology, or chemistry, to its previous model, Opus 4.8. In contrast, “Claude Mythos 5” is described as the unrestricted version of the same model, with those safeguards lifted.
For that reason, Mythos remains limited to a highly vetted group of trusted partners, such as government cyberdefenders and specific life sciences researchers.
Fable 5, meanwhile, is being offered more widely—but only briefly. Anthropic says the model is free for a limited time, and it consumes tokens faster than any other model.
That speed isn’t theoretical. In tests observed by BleepingComputer, Fable used approximately 1 million tokens in 8 minutes. The same tests found that when using Workflow—a new execution system that allows Claude to break complex prompts into smaller tasks and spin up parallel subagents—Claude Fable 5 exhausted a $100 Max subscription’s daily usage from zero in just 9 minutes.
Those results don’t show up during casual use, the reporting notes. But switch into Workflow mode and set model thinking to high, and the token burn can become immediate. Even without Workflow at “xhigh” effort, BleepingComputer observed that Fable still consumes tokens about 2 times faster than the Opus model.
Anthropic’s reluctance to unlock Fable 5 at the same capacity as Opus and its previous models ties directly to what it calls the reality of running it: it’s an expensive model because it requires a lot of compute. and that means Anthropic can’t make it available as easily as Opus 4.8 or earlier models.
The timeline makes that financial pressure tangible. Until June 22, Anthropic says Fable 5 will be offered to all Pro, Max, and Enterprise customers. After that window expires, Anthropic says it will switch Fable 5 to usage-based pricing.
There’s also a pattern Anthropic has been known for: increasing capacity later, after model “nerfs” and staged unlocks. The question now is how long customers get before the faster. pricier version turns into something harder to access—especially for anyone whose daily limits are measured in minutes when they choose to push Workflow and higher thinking settings.
Anthropic Claude Fable 5 Mythos Opus 4.8 AI models token usage Workflow cybersecurity safeguards pricing compute costs
So it shows up then just disappears? classic tech move.
I saw “safer” and assumed it would be cheaper lol. But then it’s token-hungry and only for Pro/Max/Enterprise? basically paywalling common sense.
Wait, they said Mythos was risky for hackers but now Fable 5 is the safer one from the same class. So like… does that mean it’s still good at finding vuln stuff, just with manners? Kinda sounds like the same engine with a leash.
“Countdown” is right. I swear these companies roll out stuff for a night and then throttle it the next day. Also the part about attackers vs defenders… I don’t get it, like if it’s for defenders why limit it to big paying accounts? My guess is the “guardrails” are just marketing and it’ll still eat your tokens in minutes anyway.