Anthropic Unveils Claude Fable 5 and Mythos 5: Advanced AI with Dual Safety Approaches

Anthropic has officially launched Claude Fable 5, marking a significant advancement in its suite of AI models. Released on June 9, 2026, Fable 5 is now broadly accessible and represents the company's most powerful model made available to the public. This new 'Mythos-class' model showcases exceptional performance across a spectrum of tasks, including complex software engineering, intricate knowledge work, and sophisticated vision capabilities. Its proficiency in handling longer and more complex tasks significantly surpasses previous Claude iterations, demonstrating a notable leap in AI capability. A key aspect of this release is Anthropic's innovative approach to AI safety. While Fable 5 offers cutting-edge performance, it incorporates built-in safeguards. These mechanisms are designed to detect and reroute queries related to sensitive areas such as cybersecurity, biology, chemistry, and distillation (extracting model capabilities) to a less powerful model, Claude Opus 4.8. This ensures that the advanced capabilities of Fable 5 are not misused, even if it occasionally flags harmless requests. Alongside Fable 5, Anthropic also unveiled Claude Mythos 5. This model shares the exact same underlying architecture and capabilities as Fable 5 but operates with certain safeguards removed. Mythos 5 is not available for general public use; instead, it is exclusively provided to a vetted group of cyber defenders and critical infrastructure providers through initiatives like Project Glasswing. This restricted access allows these specialized users to leverage Mythos 5's full, uninhibited power, particularly its unparalleled cybersecurity capabilities, for defensive purposes without the limitations imposed on the public version. The pricing for both Fable 5 and Mythos 5 is set at $10 per million input tokens and $50 per million output tokens, a reduction from earlier Mythos Preview models. Fable 5 is immediately available via the Claude API and is included in various subscription plans until June 22, after which usage credits will apply. Anthropic has also implemented a 30-day data retention policy for both models for safety monitoring, though this data is not used for training and is deleted after the retention period. This dual-model strategy underscores Anthropic's commitment to pushing the boundaries of AI while navigating the complex landscape of responsible and safe deployment.