TL;DR
Anthropic released Claude Fable 5 on June 9, 2026, calling it its most capable model yet made generally available. The same underlying model also exists as Claude Mythos 5, but that version has fewer safeguards in some areas and remains limited to trusted partners.
Anthropic released Claude Fable 5 on June 9, 2026, making what it calls its most capable model yet available to the public while keeping an unlocked version, Claude Mythos 5, limited to trusted partners. The release matters because Anthropic is offering a higher-capability Mythos-class model through a safety design that routes flagged requests to Claude Opus 4.8 instead of simply refusing them.
Fable 5 is the safeguarded public version of the same underlying model as Mythos 5, according to the launch material. Fable 5 ships with safety classifiers active and is available through the API as claude-fable-5. Mythos 5 is described as the same model with safeguards lifted in some areas, with access limited at first to Project Glasswing cyber-defenders and later to select biology researchers.
The safeguards monitor categories including cybersecurity, biology and chemistry, and distillation. The release material says clear requests are handled by Fable 5, while flagged requests are routed to Claude Opus 4.8. Anthropic says users are told when that routing happens, and the material states that more than 95% of requests should remain on Fable 5 while less than 5% are expected to be handed to Opus 4.8.
Early performance claims are tied to Anthropic launch material and Every, which said it had one week of early access. Every 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 source material also cites a Stripe codebase-wide migration across a 50 million-line Ruby codebase that was completed in a day instead of an estimated two months by a team.
Builders Get Power With Guardrails
For developers and companies, Fable 5 changes the buying decision from simple model ranking to deciding when long-running agent work justifies higher cost and latency. Every described the model as strongest on hard coding tasks that can be delegated end to end, while also reporting that lower-adoption users struggled to find a use for it.
The safety routing is also a product decision, not only a policy layer. If it works as described, most routine work uses the full model, while a small flagged share is moved to a weaker model with tighter limits. That could let AI providers release stronger models more broadly while limiting access in sensitive cyber or life-science areas.
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.
How The Two Models Split
Anthropic places the Mythos class above its Opus models and had previously judged that tier too risky for broad release, according to the source material. The public release pairs a safeguarded version, Fable 5, with a more open partner version, Mythos 5.
The naming reflects that split. Anthropic says Fable comes from the Latin fabula, meaning that which is told, and is close to the Greek mythos. In practical terms, the difference is not the underlying model but what each version is allowed to answer.
“the most capable model it has ever made generally available”
— Anthropic launch material
safeguarded AI chatbot
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limits Around Routing Remain Open
It is not yet clear how the routing system will behave at scale, how often benign cyber or life-science work will be moved to Opus 4.8, or how users will audit those handoffs. The launch material says the classifier is conservative, which means some legitimate requests may be routed away from Fable 5.
The independent test record is also early. Every had a week of access, and the published figures come from reported benchmarks and examples rather than broad public usage. The durability of the coding score, the Stripe migration example, and the scientific-work claims will depend on repeat testing across more tasks.
Llama Guard in Practice: Safety Classification Pipelines for LLM Responses
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
June Rollout Moves To Credits
Access is free for Pro, Max, Team and Enterprise users through June 22, 2026, according to the rollout details. After that, Anthropic plans to move to usage credits before a later standard pricing phase.
The next signals will be API usage patterns, user reports about safety routing, any change to pricing or rate limits, and whether Mythos 5 access expands beyond Project Glasswing cyber-defenders and the planned biology-research cohort.
Platform Engineering for Artificial Intelligence: Designing scalable infrastructure, data pipelines, and model lifecycle management for generative AI and agentic protocols (English Edition)
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 safeguarded Mythos-class model, on June 9, 2026. It says this is its most capable model yet made generally available.
How is Mythos 5 different from Fable 5?
The source material describes Mythos 5 as the same underlying model with some safeguards lifted. Access is limited to trusted partners, beginning with Project Glasswing cyber-defenders and later select biology researchers.
What happens when a request is flagged?
Fable 5 uses safety classifiers for areas such as cybersecurity, biology and chemistry, and distillation. Requests flagged by those systems are routed to Claude Opus 4.8, and users are told when that happens, according to the launch details.
How much does Claude Fable 5 cost?
The reported API price is $10 per million input tokens and $50 per million output tokens. The source material says that is less than half the price of Mythos Preview and around twice Opus 4.8.
Is Fable 5 mainly for coding?
Coding is where the early independent review was strongest. Every called Fable 5 the best coding model it had tested, but also reported mixed results for writing and less clear value for users without demanding tasks.
Source: Thorsten Meyer AI
