Anthropic Launches Claude Fable 5.2 and Mythos 5.2 for Advanced Technical Tasks
Artificial intelligence company Anthropic has announced two advanced AI models, Claude Fable 5.2 and Claude Mythos 5.2, aimed at handling some of the most complicated technical challenges in software engineering, scientific research, cybersecurity and hardware development. The models were introduced on November 9, marking another major development in the increasingly competitive AI industry.
Both models are built on the same underlying technology but differ in their safety restrictions and availability. Claude Fable 5.2 is designed for broader commercial and developer access, while Claude Mythos 5.2 is reserved for verified organisations working on advanced cybersecurity and life-sciences research.
Anthropic says the new models deliver improvements in coding, reasoning, scientific analysis and visual understanding. Their responses can reportedly be generated up to 30 per cent faster than the previous 5.1 versions, potentially reducing the time required for demanding technical projects.
One of the most notable demonstrations involves mathematical research. Anthropic claims that a coordinated group of approximately 1,000 AI agents using Fable 5.2 produced a machine-verifiable proof of Fermat’s Last Theorem in around 30 hours. A previous attempt using an earlier unreleased research model reportedly required 11 days.
The company has also highlighted the models’ capabilities in physics and engineering. According to Anthropic, Fable 5.2 helped investigate two previously unresolved problems associated with black-hole physics and designed three electronic devices that successfully worked during initial hardware testing.
These demonstrations suggest that Anthropic is increasingly focusing on AI systems capable of carrying out extended, multi-stage research and engineering projects rather than simply answering questions or generating code.
Anthropic has also reported significant progress in software development. Fable 5.2 performed better than its predecessor across several coding evaluations, particularly in complex programming tasks that require planning, testing and correcting errors across multiple stages. In one internal demonstration, the model coordinated AI agents to develop an operating system over approximately eight days.
Cybersecurity is another major area of improvement. Anthropic’s evaluations indicate that Mythos 5.2 can identify previously documented software vulnerabilities and assist with advanced security research. The company has therefore introduced additional access restrictions for Mythos to reduce the possibility of its capabilities being misused.
The distinction between the two models is particularly important. Fable 5.2 includes safeguards that restrict certain cybersecurity and biological research requests. Mythos 5.2 provides access to more advanced capabilities in these sensitive areas, but only through Anthropic’s verification programmes for eligible organisations and researchers.
Anthropic has also revised its pricing structure. Both models are priced at $8 per million input tokens and $40 per million output tokens, representing a reduction compared with the previous pricing of $10 and $50 respectively. An optional high-speed mode can generate responses up to 2.5 times faster, although it comes at twice the standard token price.
Claude Fable 5.2 is available through Anthropic’s platforms and major cloud providers, including Amazon Web Services, Google Cloud and Microsoft Azure. Mythos 5.2 remains subject to restricted access and organisational verification.
The launch comes amid intensifying competition between Anthropic, OpenAI and Google in the development of increasingly capable AI systems. Companies are competing not only to improve conversational abilities but also to develop models capable of independently assisting with scientific discoveries, complex programming and industrial engineering.
Anthropic has positioned its Opus 5.5 model as a suitable starting point for everyday coding and knowledge-based tasks, while recommending Fable 5.2 and Mythos 5.2 for more demanding technical assignments. This approach reflects a broader strategy of offering different AI models for different levels of computational complexity.
The company has also acknowledged challenges in evaluating advanced AI systems. Its testing indicates that the new models are more likely to recognise when they are being assessed, potentially making controlled evaluation results less representative of their behaviour in real-world environments.
The introduction of Claude Fable 5.2 and Mythos 5.2 therefore represents more than a routine software upgrade. It highlights Anthropic’s growing ambition to develop AI systems that can participate in complex research, engineering and scientific problem-solving while maintaining different levels of access and safeguards for potentially sensitive capabilities.
