Anthropic has introduced Claude Fable 5, a new Mythos class artificial intelligence model now available for general use, alongside Claude Mythos 5, a more advanced version designed for a limited group of cybersecurity and infrastructure partners. According to the company, Fable 5 represents its most capable publicly available AI model to date, with stronger performance across software engineering, scientific research, knowledge work, vision related tasks, and analytical workflows. Anthropic stated that Fable 5 surpasses previous Claude models in handling long and complex assignments while maintaining improved efficiency. At the same time, Claude Mythos 5 is being released through Project Glasswing in collaboration with the United States government for approved cyber defenders and infrastructure providers, with broader access expected through a trusted program in the future.
Anthropic said the launch of Fable 5 comes with new safeguards aimed at limiting misuse in areas such as cybersecurity, biology, chemistry, and model distillation. When the system identifies sensitive requests, users may receive responses from Claude Opus 4.8 instead of Fable 5. The company noted that these safeguards were intentionally designed to be cautious, although harmless prompts may occasionally be affected. Early figures shared by Anthropic suggest more than 95 percent of Fable 5 sessions do not trigger fallback systems, allowing users to access the model’s full capabilities in most cases. Claude Mythos 5, meanwhile, provides the same underlying architecture but removes certain cyber related restrictions for approved partners. Anthropic stated that Mythos 5 currently offers some of the strongest cybersecurity capabilities available among AI systems and has been tested for resilience against jailbreak attempts and harmful misuse scenarios.
In software engineering, Anthropic highlighted early testing results showing Fable 5 delivering significant productivity gains. Financial technology company Stripe reported that the model completed a codebase wide migration in a single day across a 50 million line Ruby environment, a task that otherwise would have required a team working for more than two months. The model also achieved strong scores in software testing evaluations, including Cognition’s FrontierCode benchmark. Beyond engineering, Anthropic said Fable 5 demonstrated stronger performance in document reasoning, financial analysis, chart interpretation, and problem solving. The company referenced evaluations from Hebbia and IMC, which showed improved outcomes in senior level reasoning and trading analysis tasks. Fable 5 also delivered stronger vision capabilities, including extracting numerical data from scientific charts and recreating software code from screenshots with limited assistance.
Anthropic further emphasized the role of Claude Mythos 5 in scientific and biomedical research. Internal teams reportedly used the model to accelerate elements of drug development by nearly ten times, with Mythos 5 independently handling tasks such as selecting protein binding sites, operating design tools, and managing failed processes without human support. Researchers also reported that the model generated scientific hypotheses in molecular biology that were preferred over earlier Claude systems in most comparisons, with several concepts advancing to experimental review. In genomics, Mythos 5 reportedly conducted more than a week of autonomous research involving millions of cells across 138 animal species, ultimately creating a machine learning system that outperformed a recently published academic model despite using significantly fewer resources. Fable 5 is currently available through Claude subscription plans and the Claude API at pricing starting from $10 per million input tokens and $50 per million output tokens, while Mythos 5 access remains limited to approved partners and selected biology researchers.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
