Anthropic releases Claude Fable, a Mythos-class model

Anthropic launched Claude Fable 5, a new Mythos-class model, and published a companion Claude Mythos 5 configuration in a June 9 announcement. According to Anthropic's blog post, Fable 5 "shows exceptional performance in software engineering, knowledge work, and vision," and the company said the model's advantage increases on longer, more complex tasks. Anthropic's announcement describes targeted safeguards that block responses in specific high-risk domains such as cybersecurity, biology, chemistry, and distillation, and states that when safeguards trigger the system falls back to Claude Opus 4.8. Anthropic reported in its post that 95 percent of Fable sessions in testing produced only Fable responses, without fallback to Opus 4.8. The company also published a system card for Claude Fable 5 and Claude Mythos 5 with additional technical notes. Anthropic imposed a mandatory 30-day data retention policy on all Fable and Mythos traffic -- including enterprises that previously had zero-retention agreements -- framing the requirement as a safeguard against novel jailbreaks; Anthropic stated the retained data will not be used for training (TechCrunch).

Editorial analysis - technical context: Public material describes Fable 5 and Mythos 5 as the same underlying model architecture exposed under two configurations: one with stricter, built-in safeguards (Fable) and one with fewer constraints in some areas (Mythos), per Anthropic's announcement and system card. Anthropic highlights benchmark and red-team metrics in the system card (named metrics include Firefox, OSS-Fuzz, CyberGym, and CyScenarioBench) to quantify exploit and jailbreak performance. The company also references Project Glasswing, an earlier restricted-access program, as part of how it managed higher-risk Mythos capabilities before broad release. Anthropic reported running an external bug bounty that "produced no universal jailbreaks in over 1,000 hours of testing" and engaged additional external red-teaming organizations with the same result, per TechCrunch. Third-party testing by analytics company Hex found Fable 5 was the first model to achieve 90% on its core analytics benchmark of complex, long-running analytical tasks (TechCrunch).

Pricing for both Fable 5 and Mythos 5 is $10 per million input tokens and $50 per million output tokens -- double the price of Opus 4.8 (TechCrunch; CNBC). Through June 22, Fable 5 is included in Pro, Max, Team, and seat-based Enterprise plans at no additional cost. From June 23, Anthropic will remove Fable 5 from subscription plans and require usage credits going forward, with plans to restore it as a standard subscription feature as soon as possible (TechCrunch).

Public reporting frames this as a notable step in making higher-capability models more widely available while attempting to limit high-risk outputs through automated safeguards. Mythos was previewed in April in a limited partner program; on June 2, Anthropic expanded access to hundreds of organizations across 15 countries managing critical infrastructure. The current Fable 5 release is the first time a Mythos-tier configuration has been made available to anyone through the API and consumption-based Enterprise plans (TechCrunch). For practitioners, the release illustrates how vendors are packaging capability levels and safety configurations as separate access tiers and highlights continuing trade-offs between capability, safety controls, and access.

Monitor empirical evaluations of Fable 5 on security and bio-risk red-team benchmarks, and independent tests for universal jailbreaks. Watch for details in the published system card and third-party audits that quantify how often safeguards block harmful outputs and how fallback behavior affects latency and user experience. Also watch for how enterprises and security teams integrate Fable 5 versus restricted Mythos configurations, and for updates from Anthropic about access policies or changes to the safeguard layer. The mandatory 30-day retention policy may set a precedent for how access to frontier models comes bundled with safety monitoring obligations.