Anthropic’s Claude Fable 5 is a version of Mythos the public can access today
Anthropic has released Claude Fable 5, a publicly accessible version of its powerful Mythos AI model, with built-in safety limits. The model excels in software engineering, knowledge work, and vision, but defers to an older version for high-risk domains like cybersecurity. Access will transition to a credit-based system after an initial free period for subscribers.
Anthropic has made its advanced AI model, Claude Fable 5, available to the public for the first time. This release marks the public debut of a version of their powerful Mythos model, offering advanced capabilities for various tasks. The company emphasizes the integration of robust safety measures within the model.
Fable 5 is designed to excel in areas such as software engineering, knowledge work, and vision. However, to mitigate risks, it incorporates strict safety limits. In highly sensitive domains like cybersecurity, biology, chemistry, and distillation, the model will automatically defer to an older version, Claude Opus 4.8, to prevent potential misuse or errors.
Initially, the Mythos model was restricted to a select group of partners due to cybersecurity concerns. Anthropic later expanded access to hundreds of organizations globally, focusing on those managing critical infrastructure. The public release of Fable 5 through Anthropic's Claude API and Enterprise plans signifies a broader accessibility strategy.
Access to Fable 5 will roll out in stages, with an initial complimentary period for subscribers. Subsequently, usage will require credits, aligning with Anthropic's pricing structure of $10 per million input tokens and $50 per million output tokens, which is double the cost of Opus 4.8.
Anthropic conducted extensive stress testing and external bug bounties to identify and address potential vulnerabilities before the public launch. Despite these efforts, the company acknowledges that novel attacks may still be possible. As a result, Anthropic will implement a mandatory 30-day data retention policy for all traffic, exclusively for defensive purposes and not for model training.
Third-party evaluations highlight Fable 5's exceptional performance. Hex, an analytics company, noted its 90% score on complex analytical tasks, praising its judgment and attention to nuance. Similarly, Base44 and Genspark lauded Fable's ability to develop full applications and outperform other models in UI design and game coding.
Related articles
Build real agentic apps using CUGA: two dozen working examples on a lightweight harness
CUGA, IBM's open-source Agent Harness, simplifies building agentic applications by handling infrastructure, allowing developers to focus on tools and prompts. It offers pre-assembled components for planning, execution, and state management, significantly reducing development time. CUGA has topped agent benchmarks like AppWorld and WebArena.
OpenAI launches new initiative to help find and patch open source bugs
OpenAI has launched "Patch the Planet," a new initiative in partnership with cybersecurity firm Trail of Bits, to enhance the security of open-source projects. This program aims to assist maintainers in identifying and patching bugs, utilizing OpenAI's AI-powered security tools while reducing the burden on project teams.
PP-OCRv6 on Hugging Face: 50-Language OCR from 1.5M to 34.5M Parameters
Baidu has released PP-OCRv6, an advanced optical character recognition (OCR) model supporting 50 languages. Available on Hugging Face, this version significantly improves accuracy and efficiency across various parameter sizes, from 1.5 million to 34.5 million, marking a substantial leap in multilingual OCR technology.
