Table of Contents
ToggleThe global AI ecosystem is witnessing a historic leap in efficiency. On June 9, 2026, Anthropic, a leader in artificial intelligence, officially announced its new generation "Mythos-class" top-level model architecture, and simultaneously launched Claude Fable 5, which is available to the public, and Claude Mythos 5, which is limited to specific defense applications.
The official statement said that this series of models has set new state-of-the-art (SOTA) benchmark records in fields such as software engineering, visual code reconstruction, long text deep memory, and life science research, and that "the longer and more complex the task, the more significant its leading advantage becomes."
Amazing efficiency: A two-month project completed in one day, AI-driven drug design accelerated 10 times.
In the officially released capability assessment, Claude Fable 5 / Mythos 5 demonstrated cross-generational autonomous task handling capabilities:
- Software Engineering: In Stripe's actual testing, the model significantly reduced the complex migration work, which originally required the entire engineering team to spend two months and involved 50 million lines of Ruby code, to a "one-day" refactoring.
- Multimodal vision: Reaching the next-generation SOTA level, it can not only reconstruct the code 100% from a webpage screenshot, but also play the classic game Pokémon FireRed independently without any additional assistance, relying solely on "original screen input".
- Life Sciences (Mythos 5): In collaboration with the US government and scientists, Mythos 5 has successfully increased the speed of drug design by 10 times, enabling it to autonomously design proteins and propose novel scientific hypotheses. Its blind test preference rate among scientists is as high as 80%.
It features an innovative "automatic downgrade" mechanism, switching to Opus 4.8 for high-risk queries.
With the release of god-like abilities, the risk of AI being misused has reached unprecedented levels. To prevent extreme risks such as exploitation of powerful cybersecurity vulnerabilities and synthesis of biological weapons, Anthropic pioneered a dynamic "automatic downgrade" security barrier in the publicly available version of Fable 5.
When the built-in dedicated AI classifier detects that a user request involves highly sensitive cybersecurity attacks, biological/chemical hazards, or malicious model distillation, the system automatically downgrades the conversation on the device or in the cloud, replacing it with a safe and compliant response from the secondary model, Claude Opus 4.8. Official data indicates that this conservative safeguard only affects less than 5% of conversations on average, and the system proactively notifies the user. Future optimizations will continue to reduce the false positive rate.
In partnership with the US government's Project Glasswing, pricing was revealed simultaneously.
Compared to Fable 5, which has a security lock, Claude Mythos 5, with its core unlocked, demonstrates its ultimate defensive capabilities. Currently, Mythos 5 is only strictly limited to network defenders and critical infrastructure providers participating in Project Glasswing (a critical infrastructure protection project in cooperation with the US government).
Regarding availability and pricing, the public version of Fable 5 is now officially available to users worldwide (including the webpage, app, and developer API). The API is priced at $10 USD per million tokens input and $50 USD per million tokens output , less than half the price of the previous Mythos beta version, making it highly competitive. Furthermore, the official announcement states that from June 9th to June 22nd, existing Pro, Max, Team, and Enterprise subscribers will have the opportunity to experience this new Mythos-level cutting-edge AI capabilities for free.
Related reports
