Anthropic has released Claude Fable 5, its most powerful AI model yet made available to the general public. Built on the same underlying technology as its restricted Mythos model class, it brings frontier-level AI capabilities to everyday users — with carefully designed safety limits baked in. Access is free for subscribers until June 22, after which usage-based pricing kicks in.
What Is Claude Fable 5?
To understand what this model is, it helps to know what came before it. Earlier this year, Anthropic quietly launched a model called Claude Mythos Preview — an AI system so capable that the company judged it too risky to release publicly. Its skills in cybersecurity, biological research, and software engineering were advanced enough that Anthropic worried it could give bad actors a meaningful advantage. Mythos Preview went to vetted government cyberdefenders and critical infrastructure providers only, through a program called Project Glasswing.
Fable 5 is built on that same Mythos-class foundation. The difference is a new set of safety classifiers — separate AI systems that monitor incoming requests. When a query touches sensitive areas like offensive cybersecurity, biology, or chemistry, the classifier routes that request to Claude Opus 4.8 instead. The user still gets a response; it just comes from a slightly less capable model. Anthropic says this fallback happens in fewer than 5% of sessions on average.
The name is deliberate. “Fable” comes from the Latin fabula, meaning “that which is told” — a nod to its relationship with Mythos, from Greek, also meaning “story.” The safeguards are what distinguish the two models.
What the Model Can Do
Fable 5 is a significant step up from previous Claude models. In software engineering, early testing showed it compressing months of development work into days. One company completed a migration across a 50-million-line codebase in a single day — a task that would have taken a full engineering team more than two months.
It also performs strongly on long-context tasks, maintaining focus and accuracy across millions of tokens. That matters for complex, multi-step work like legal document review, financial analysis, or extended research projects. Partners in early testing described it as the first model to handle senior-level analytical tasks with genuine, consistent reliability.
Vision capabilities have also improved noticeably. The model can extract precise data from detailed scientific figures, rebuild web application source code from screenshots alone, and handle complex visual reasoning without the extra scaffolding that earlier models required.
The Safety Tradeoffs
Anthropic is being unusually transparent about the limitations of its safeguards. Because the classifiers run on the cautious side, they will sometimes catch legitimate requests. A biology researcher asking a routine question could find themselves redirected to Opus 4.8. Anthropic acknowledges the frustration and says it is actively working to cut down on false positives.
The cybersecurity safeguards are the strictest. The model will not assist with exploit development, offensive hacking tasks, or cyberattack planning. In external testing, it refused every harmful single-turn request in those categories, even when testers applied 30 different known jailbreak techniques. More than 1,000 hours of red-team testing produced no universal jailbreaks.
Alongside the launch, Anthropic updated its data handling policy for Mythos-class models. All traffic on these models now goes into a 30-day retention window. The company will not use that data to train new models, but it will draw on it to detect novel attack patterns and reduce false positives. Access to retained data is logged, and the data is deleted after 30 days in almost all cases.
How Fable 5 Compares to Mythos 5
Claude Mythos 5 launched on the same day and shares the same underlying model. The key difference is that Mythos 5 drops the cybersecurity safeguards for users who qualify. Access stays restricted to Project Glasswing partners, including US government cyberdefenders, with a broader trusted access program in the works. A separate program for biology researchers is also in development.
Both models carry the same pricing: $10 per million input tokens and $50 per million output tokens. That is less than half the cost of Claude Mythos Preview.
Availability and Token Usage
Fable 5 is available now across all Claude plans. Pro, Max, Team, and Enterprise subscribers can use it at no extra cost until June 22. After that date, usage requires credits. Anthropic plans to bring it back as a standard subscription feature once capacity allows, but has not committed to a specific date.
One practical thing to know: the model consumes tokens at roughly twice the rate of Opus in normal use. Switch on Workflow mode — which splits complex tasks across parallel subtasks — and usage climbs much faster. Users on fixed subscription plans should keep that in mind, especially during the free access window.
Final Thoughts
Fable 5 marks a real shift in what consumer-facing AI can do. Anthropic spent months figuring out how to bring Mythos-class capabilities to a general audience without handing powerful tools to bad actors. The result is a model that genuinely impresses — but one that carries intentional limits, honest caveats, and a pricing change on the horizon. If you are on a Pro or Max plan, the free access window closes on June 22.
