TL;DR
Anthropic released Claude Fable 5 on June 9, 2026, calling it its most capable generally available model. The release matters because Fable 5 is a public version of the same underlying model as Mythos 5, with safety classifiers that route flagged prompts to Claude Opus 4.8 instead of refusing them.
Anthropic released Claude Fable 5 on June 9, 2026, making what it calls its most capable model generally available while using a new safeguard system that routes risky requests to the weaker Claude Opus 4.8 rather than issuing a direct refusal.
According to the source material, Fable 5 and Mythos 5 are the same underlying model. Fable 5 is the public, safeguarded version, while Mythos 5 is the less restricted version limited first to Project Glasswing cyber-defenders and expected later for select biology researchers.
Anthropic describes Fable 5 as a Mythos-class model, a tier it places above Opus. The company previously treated that class of capability as too risky for broad release, according to the material. In the public version, safety classifiers monitor areas including cybersecurity, biology and chemistry. If a prompt is cleared, Fable 5 responds with full capability. If flagged, the request is routed to Claude Opus 4.8, and the user is told that routing occurred.
The company’s reported price is $10 per million input tokens and $50 per million output tokens, less than half the price cited for Mythos Preview and about twice the price of Opus 4.8. The model is described as free on Pro, Max, Team and Enterprise plans through June 22, followed by usage credits before standard billing later.
A New Safety Model
The release is less about a single benchmark result than about a new way to ship frontier capability. Anthropic is allowing public access to a model it classifies above Opus while limiting higher-risk uses through routing, not blanket denial.
That matters for developers and companies because it changes how powerful AI systems may be sold and governed. If the approach works as described, users can get full model performance for most tasks while the provider limits certain misuse paths. If the classifiers are too broad, legitimate security or life-science work may be pushed to a weaker model, changing the quality of results for exactly the users who may need advanced capability most.
The economics also matter. At the reported price, Fable 5 is positioned for larger, higher-value jobs rather than routine chat or quick edits. The source material frames the model as best suited to asynchronous agent work: code migrations, long-running engineering tasks, app generation and other jobs where time savings can outweigh model cost.
Portable AI Language Translator Device with Photo Recognition for Real Time Bidirectional Voice Translation – 139 Online and 20 Offline Languages, WiFi Connectivity and 1500mAh
Multilingual Online and Offline Support: This device provides translation for 139 languages and dialects online and 20 languages…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Opus To Mythos
Anthropic’s prior public flagship tier has been Opus. Fable 5 introduces a public Mythos-class release, while Mythos 5 remains restricted. The name distinction, according to the source material, reflects access controls rather than a separate base model.
The evidence cited in the material includes a reported Stripe codebase-wide migration across a 50 million-line Ruby codebase completed in one day instead of an estimated two months by a team. Every, an independent reviewer given a week of early access, scored 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.
Other claims attributed to the launch material include state-of-the-art vision performance, screenshot-to-code reconstruction, completion of Pokemon FireRed using a vision-only harness, drug-design acceleration with Mythos 5, and a genomics model trained by Mythos 5 that reportedly beat a recent Science result at one-hundredth the size. These are source-reported claims and have not been independently verified here.
“The best coding model in the world”
— Every, early-access reviewer
safeguarded AI chatbot
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Open Questions On Routing
Several details remain unsettled. It is not yet clear how often benign cybersecurity, biology or chemistry prompts will be routed to Opus 4.8 in regular use. The source material says flagged prompts are expected to be less than 5 percent and that the classifier is tuned conservatively, but real-world rates may vary by user and domain.
It is also unclear how Anthropic will handle appeals, auditability, enterprise policy controls and cases where a user needs the stronger model for legitimate high-risk research. The source material also flags 30-day retention as a compliance concern, but does not provide full policy detail.
The largest capability claims also need broader testing. Early access results, company examples and reviewer benchmarks can be useful signals, but they do not settle how Fable 5 performs across different codebases, scientific workflows, languages, safety edge cases or cost-sensitive production settings.
Uncensored & Underground: Black Hat ChatGPT AI and Jailbreaks: While Mainstream LLM Models Say “No,” a Parallel Market is Busy Teaching Models to Say “Yes”—to … AI: The Black Hat ChatGPT Series Book 8)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Rollout Moves To Credits
The next milestone is the end of the free access window on June 22, 2026, for Pro, Max, Team and Enterprise users, followed by usage credits and later standard pricing. That schedule will give developers and businesses an early test period before cost becomes a larger factor.
Anthropic is also expected to expand Mythos 5 access beyond Project Glasswing to select biology researchers, according to the source material. The pace and terms of that access will show how far the company is willing to extend the less restricted version while keeping the public release behind classifiers.
Behavioral Cybersecurity
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Anthropic announce?
Anthropic announced Claude Fable 5, a generally available Mythos-class model, and described Claude Mythos 5 as the same underlying model with some safeguards lifted for trusted partners.
How is Fable 5 different from Mythos 5?
According to the source material, the model itself is the same. Fable 5 includes public safeguards, while Mythos 5 is less restricted and limited to approved users such as Project Glasswing cyber-defenders.
Does Fable 5 refuse risky prompts?
The source material says Fable 5 routes flagged prompts to Claude Opus 4.8 rather than refusing them outright. Users are told when that routing happens.
Why are developers paying attention?
Early reports describe strong performance on long coding tasks, including Every’s 91 out of 100 Senior Engineer benchmark score. The model may be most useful for large, well-defined work where higher cost and latency can be justified.
What remains unverified?
Broader independent testing is still needed on cost, reliability, classifier accuracy, scientific claims, enterprise controls and real-world performance outside early access examples.
Source: Thorsten Meyer AI
