TL;DR
Anthropic released Claude Fable 5 on June 9, 2026, calling it its most capable generally available model. The same underlying model is offered as Claude Mythos 5 to selected trusted partners with fewer safeguards, while public Fable 5 routes flagged cyber, biology and chemistry requests to Claude Opus 4.8.
Anthropic released Claude Fable 5 on Tuesday, June 9, 2026, making what it calls its most capable generally available Claude model available to users while keeping the less restricted Claude Mythos 5 version limited to trusted partners. The launch matters because Anthropic is putting a model class it previously treated as too risky for broad release into public use through a safety system that routes flagged requests to a weaker model instead of issuing a blanket refusal.
Anthropic describes Fable 5 and Mythos 5 as the same underlying model. The difference, according to the source material, is access and safeguards: Fable 5 is the public version with safety classifiers active, while Mythos 5 has some of those limits lifted and is restricted first to Project Glasswing cyber-defense partners, with select biology researchers expected later.
On clear requests, Fable 5 answers with the full model. On requests flagged in areas such as cybersecurity, biology, chemistry or model distillation, the system sends the prompt to Claude Opus 4.8. The source material says fewer than 5% of requests are expected to be routed this way and that users are told when it happens, though conservative classifiers may catch benign work.
Anthropic prices Fable 5 at $10 per million input tokens and $50 per million output tokens, less than half the price of the prior Mythos Preview program but about twice Opus 4.8. The model is available through Claude plans and the API under claude-fable-5.
Power Moves Behind Guardrails
For developers and companies, the release changes how frontier models may be bought and used. Anthropic is offering high capability by default and using a fallback model when classifiers detect higher-risk work. That design could shape how labs make powerful systems available without giving all users the same level of access.
The reported performance claims also point to a shift in how some teams may use Claude. Every, which said it had a week of early access, rated Fable 5 at 91 out of 100 on its Senior Engineer benchmark, compared with 63 for Opus 4.8 and 62 for GPT-5.5. The same review called it strongest on long, difficult, well-scoped engineering tasks, while less useful for casual or poorly specified work.
The cost profile matters too. At roughly twice the price of Opus 4.8 and more than three times Sonnet 4.6, Fable 5 is positioned less as a routine chat model than as a tool for larger jobs where the user can justify the wait, token use and review burden.
Prompt Engineering for LLMs: The Art and Science of Building Large Language Model-Based Applications
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Mythos Moves Into Public Use
The Mythos class had been associated with Anthropic’s most restricted frontier systems before this release. In the supplied materials, Anthropic is described as having judged earlier Mythos-level capability too dangerous for broad access, especially because of cybersecurity and life-science misuse risks.
The naming reflects the product split. Anthropic says Fable derives from Latin fabula, meaning that which is told, close to the Greek mythos. The names mark the access layer: Fable 5 is the public version, while Mythos 5 is the same model offered with fewer limits to a smaller set of approved users.
“most capable model it has ever made generally available”
— Anthropic launch materials
Artificial Intelligence for Cybersecurity: Develop AI approaches to solve cybersecurity problems in your organization
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Classifier Effects Remain Unproven
Several points are not yet clear. Anthropic’s routing system is described as affecting fewer than 5% of requests, but the source material does not give an independent audit of false positives, false negatives or how often legitimate security and life-science work is sent to Opus 4.8. It is also unclear how much appeal or override capacity paying users will have when benign work is rerouted.
The performance evidence is still developing. The Every benchmark and company examples point to large gains, including a reported Stripe migration across a 50 million-line Ruby codebase in one day and a 91 out of 100 coding score. Those figures are attributed to the sources that reported them; the supplied material does not include raw benchmark data, full methods or third-party replication.
AI-Powered Bioinformatics for Wet Lab Researchers: Dry Analysis Through Dialogue Alone
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Credits Begin After June 22
The next milestone is the end of the free-access window on June 22, 2026, after which Anthropic says Fable 5 use will move to credits unless capacity allows the company to extend the period. Standard billing is expected later, making usage patterns and model-routing behavior the immediate items for teams to monitor.
For Mythos 5, the next phase is controlled expansion. Anthropic is keeping the less restricted model with trusted partners for cyber-defense work and plans to add selected biology researchers.
Building Web Apps with Bolt: Unlock the Future of AI-First Development
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Anthropic release?
Anthropic released Claude Fable 5, a public Mythos-class Claude model, on June 9, 2026. The company says it is its most capable generally available model.
Are Fable 5 and Mythos 5 different models?
The source material says they are the same underlying model. Fable 5 has public safeguards active; Mythos 5 is offered to a smaller set of trusted users with fewer limits in some areas.
What happens when Fable 5 detects a risky request?
The system routes flagged requests to Claude Opus 4.8 instead of having Fable 5 answer them. The material says users are told when routing occurs and that the filter may catch benign requests.
How much does Fable 5 cost?
The listed API price is $10 per million input tokens and $50 per million output tokens. That is about twice Opus 4.8 and less than half the prior Mythos Preview price, according to the supplied material.
Who can use Mythos 5?
Access is limited to Project Glasswing cyber-defense partners at launch, with selected biology researchers expected later. Broad public access to Mythos 5 has not been announced.
Source: Thorsten Meyer AI
