For three days in June 2026, you could use the most capable public AI model ever released. Then it was gone.
Claude Fable,
before it got banned
The record of what Claude Fable was, what it was genuinely good at, and how we used it before a government directive took it offline. Plus the business lesson that matters more than the model itself.
72h
Live, then gone
9 Jun
Launched
+10pts
Beat Opus 4.8
12 Jun
Pulled offline
Timeline
What Fable Actually Was
The public release of Mythos, and the three days it existed.
9 June 2026
Fable 5 goes public
The public release of Anthropic's Mythos-class model. Free on Pro, Max, Team, and Enterprise plans.
Planned: 23 June
Usage-based pricing
Anthropic called it its most capable public model, beating Opus 4.8 by more than 10 points on several benchmarks. The pricing transition never happened.
12 June 2026
Taken offline
A US Department of Commerce export-control directive took it offline. Three days after launch.
The full timeline is in Claude Fable: Rise and Fall.
Capability
What It Was Genuinely Good At
Where Fable pulled ahead of Opus 4.8, from hands-on use.
Fable was a real step up from Opus. The clearest gains were on the hard, structural work.
Complex reasoning
Held long, tangled reasoning together and got more of it right in a single pass.
Framework logic
The logic behind my AI frameworks, worked out deeper than Opus managed.
Mapping at scale
Mapping every item in my internal wiki, cleanly and in one go.
Dashboards
Pushed my dashboards to a deeper level than Opus reached.
Not everyday content, where the difference was small. The jump showed on the dense logic, large maps, and reasoning-heavy tasks where a weaker model quietly drops quality.
In practice
How We Used It at Orchestration Level
Fable on top, delegating to Opus agents underneath.
I moved fast and used Fable to improve my dashboards and map my frameworks properly. The pattern that worked: run Fable at the orchestration level, then delegate execution to cheaper Opus agents underneath.
Fable on top
Hard reasoning and direction
Opus agent
Opus agent
Opus agent
Cheaper execution underneath
It is the same shape we teach in the anti-fragile AI business pillar: a smart layer on top, swappable workers underneath, and the method owned separately from any one model.
Enjoying the guide?
Enter your details to unlock the remaining insights. We'll also send you the complete guide as a reference.
The event
Why It Got Pulled
The export-control directive, stated factually.
The directive
On 12 June 2026, the US Department of Commerce cited national-security concerns tied to a reported narrow jailbreak of the Mythos 5 model.
Disabled globally
Anthropic disabled both Fable 5 and Mythos 5 for all customers the same day.
AWS revoked access
Access was pulled in parallel across cloud providers. No notice, no transition.
Anthropic publicly disagreed and said it was working to restore access under modified safeguards. Other Claude models stayed unaffected. The point is not the politics. It is that any single model can be switched off with no notice.
The takeaway
The Lesson: Own the Method
What Fable teaches every coach, consultant, and cohort owner.
Fable proved the rule in 72 hours. You may not own the model, but you can own the method.
If your delivery stopped when Fable went offline, you built a dependency, not a system.
The fix is not to chase more models
Extract your method
Keep it portable
Keep the model swappable
Keep a human on the decisions that matter
That is owning your AI stack.