You guys — this is one of those stories worth pinning! Anthropic just announced Claude Fable 5, their first "Mythos-class" model ever made widely available after previously calling the class too dangerous for public release due to cyber capabilities. They claim it excels at software engineering, knowledge work, and vision, with performance gap over rivals widening as tasks get longer and more complex. The company says this rollout was only possible thanks to new guardrails that block high-risk responses; in testing, 95% of Fable sessions ran entirely on the model without falling back to Claude Opus 4.8 (the "honest" model from last month), which is where it routes for sensitive content. They also released Mythos 5 — same base as Fable but with fewer safeguards — restricted currently to organizations in their Project Glasswing program, and they're promising a broader trusted-access rollout over time.

And the pricing story tells you everything about how serious this tier is: both models cost $10 per million input tokens and $50 per million output tokens, which doubles Opus 4.8 rates but sits at half the price of Mythos Preview — so they're positioning Fable as their accessible flagship while keeping the unrestricted version behind gates. Side note that Anthropic didn't explain why these are numbered "5" when there was no prior Fable or Mythos series, nor how they relate to Claude Mythos Preview; they simply declined comment on those specifics at The Verge’s request. That says something about where this fits in their roadmap without them having to say it outright.

Source: https://www.theverge.com/news/946725/anthropic-releases-claude-fable