Overview
- Anthropic launched Claude Fable 5 as a public, safety‑mitigated version of its Mythos‑class model that blocks or reroutes high‑risk queries in areas like cybersecurity, biology, chemistry, and model distillation to the older Claude Opus 4.8.
- Claude Mythos 5, the less‑restricted build, remains available only to vetted cyberdefense and critical‑infrastructure partners through Project Glasswing, while Anthropic says it will expand trusted access over time.
- The company set new pricing at $10 per million input tokens and $50 per million output tokens, imposed a mandatory 30‑day retention window for Mythos‑class traffic for safety monitoring, and staged subscription access through June 22 before switching to usage credits on June 23.
- Anthropic says it ran over 1,000 hours of internal and external red‑teaming and a bug bounty before launch with no universal jailbreak found, but users and researchers criticized the model for quietly downgrading some requests and the company has updated Fable 5 to make fallbacks visible and to return refusal reasons on the API.
- The move, announced June 9 and updated June 11 with transparency changes, aims to let defenders use frontier capabilities at scale while trying to limit misuse, and it comes as Anthropic manages demand, compute partnerships, and a confidential IPO process that shape pricing and access decisions.