Anthropic launched Claude Fable 5, its first public Mythos‑class model. The company says Fable 5 surpasses every previously released Claude model in software engineering, knowledge work, vision and scientific research, and beats Opus models on long, complex tasks. It can run autonomously longer than any prior Claude, marking a step toward more capable general‑purpose AI. Its performance on multimodal inputs also eclipses prior releases.

To curb misuse, Anthropic layered conservative safeguards. About five percent of sessions trigger a fallback to Opus 4.8, especially queries involving cybersecurity, chemistry or biology. A separate offering, Claude Mythos 5, reaches a limited group of cyberdefenders and infrastructure providers with fewer restrictions, positioned as the world’s strongest AI for security work via Project Glasswing. Mythos 5 also integrates real‑time threat intel feeds for faster response.

Pricing drops to $10 per million input tokens and $50 per million output tokens, under half the cost of the Claude Mythos Preview. Fable 5 is bundled in Pro, Max, Team and Enterprise plans through June 22; after that date users must purchase usage credits. Anthropic aims to restore the model to subscription plans once capacity stabilizes. Early adopters report noticeable speed gains in code generation tasks.