Claude Fable 5.1 and Mythos 5.1 Launch With Lower Costs and Stronger Safeguards

Anthropic has officially launched Claude Fable 5.1 and Claude Mythos 5.1, marking a significant advancement in large language model capabilities for coding, knowledge work, and agentic scientific research. Both models share identical underlying architectures but differ in their safety configurations. Fable 5.1 is publicly available across all major cloud platforms and the Claude API, while Mythos 5.1 maintains stricter access controls tailored for cybersecurity professionals and life sciences researchers. The new models deliver substantial performance gains across multiple benchmarks, particularly in long-horizon problem-solving and autonomous task execution. In practical applications, Fable 5.1 has demonstrated the ability to resolve complex, multi-year software crashes and autonomously optimize deep learning GPU kernels, cutting research compute costs by up to sixty percent. Anthropic also highlighted early scientific breakthroughs enabled by Mythos 5.1, including the generation of high-affinity protein binders with hit rates nearly five times higher than industry averages, and the creation of high-resolution elevation maps of Venus using historical NASA radar data. Alongside performance improvements, Anthropic has addressed longstanding enterprise concerns regarding pricing and data privacy. The company has reduced cache read costs by seventy-five percent, translating to an approximate twenty-five percent reduction in overall workload expenses, with savings reaching forty-five percent for highly agentic tasks. To enhance data sovereignty, Anthropic introduced Enterprise Frontier Safeguards, a privacy-first system that stores all customer data on client-controlled cloud infrastructure rather than Anthropic servers. This rollout, beginning this autumn, enables true zero data retention policies without compromising advanced misuse detection. Safety and compliance frameworks have been rigorously updated to match the models expanded capabilities. Safeguards for cybersecurity and biology tasks now generate sixty percent fewer false positives, allowing defensive vulnerability scanning while maintaining strict controls on exploit generation. The models also feature strengthened anti-distillation mechanisms, including restrictions on manually editing conversation history to prevent capability extraction. Furthermore, in compliance with upcoming European Union regulations, Claude outputs will incorporate an invisible numerical watermark and a detection API for authorized entities to verify AI-generated content. Claude Fable 5.1 is immediately available to developers and enterprise clients through the standard API and major cloud providers at $10 per million input tokens and $50 per million output tokens. Claude Mythos 5.1 remains accessible exclusively through vetted trusted access programs, currently prioritizing U.S.-based cybersecurity and biomedical organizations in partnership with federal authorities. The launch signals a strategic push toward integrating advanced AI directly into scientific discovery and enterprise infrastructure, with continued emphasis on safety, cost efficiency, and regulatory alignment.