What Claude Fable 5 Is and Why It Matters
Claude Fable 5 is Anthropic’s first publicly accessible Mythos-class AI model, designed to bring frontier AI capabilities such as advanced coding, reasoning, and tool use to a broad audience while keeping tighter controls on its most sensitive behaviors. Built on the same Mythos-class architecture as Claude Mythos 5, Fable 5 exposes much of the performance of Anthropic’s top frontier AI models through the Claude API and subscription plans, but with extra limits on cybersecurity and bioscience content. This public release marks a shift from the earlier Mythos preview, which was restricted to a small set of partners because of its powerful cyber reasoning skills and vulnerability discovery. By offering a guarded variant of its strongest system to everyday developers and knowledge workers, Anthropic is testing how far it can democratize Mythos-level performance without giving up its safety posture.
Mythos-Class Performance, With Guardrails Built In
Both Claude Fable 5 and Claude Mythos 5 share the same underlying model and pricing, and their benchmark numbers place them firmly among frontier AI models. On SWE-Bench Pro, they reach 80.3%, beating Anthropic’s earlier Claude Mythos Preview at 77.8 and Claude Opus 4.8 at 69.2, and they score 29.3% on FrontierCode Diamond, more than double Opus 4.8’s 13.4. Anthropic says Fable 5 is particularly strong at software engineering, knowledge work, vision, and agentic tasks, posting 17.4% on AutomationBench and 85.0% on OSWorld-Verified. Yet Fable 5 contains explicit AI safety guardrails: when prompts touch on cybersecurity, biology, chemistry, or distillation, the system refuses and hands the request to Claude Opus 4.8 instead. Early telemetry suggests that “at least 95% of Fable sessions run entirely on the model’s own answers,” so most users still see full Mythos-class AI behavior.
Tiered Access: Mythos 5 for Critical Users, Fable 5 for Everyone Else
Anthropic’s launch strategy splits Mythos-class AI into two tiers. Claude Mythos 5 is the most capable and least restricted version, with broader access to the cyber reasoning skills that made the preview controversial. Rather than a general roll-out, Mythos 5 is delivered through Project Glasswing to selected cybersecurity firms, infrastructure operators, and strategic partners, now spanning more than 15 countries. Claude Fable 5 is the public-facing counterpart, exposing the same core architecture but under stricter content rules. Both models carry the same usage-based pricing of USD 10 (approx. RM46) per million input tokens and USD 50 (approx. RM230) per million output tokens, double Claude Opus 4.8’s rate. Fable 5 is temporarily bundled into Pro, Max, Team, and seat-based Enterprise plans at no extra charge before shifting to usage credits, positioning it as the entry point to Mythos-class AI while Mythos 5 stays gated.
Data Retention and Safety as a Condition of Power
To ship Mythos-class AI at scale, Anthropic is tying access to stricter safety processes and data rules. Before releasing Claude Fable 5, the company ran extensive red teaming, saying an internal bug bounty found “no universal jailbreaks in over 1,000 hours of testing,” followed by external teams that also failed to break its classifiers universally. Even with those results, Anthropic now enforces mandatory 30-day data retention for all traffic going through Fable 5 and Mythos 5, including enterprises that previously held zero-retention contracts. The firm says it will not train on this data and will use it only to detect complex, novel attacks and reduce mistaken refusals. This trade-off makes Mythos-class AI usable for more customers, but at the cost of stricter data policies that frame storage as a safety tool rather than a product analytics feature.
Balancing Democratization, Monetization, and Market Position
Claude Fable 5 arrives as Anthropic prepares for an IPO and competes head-to-head with other labs offering frontier AI models. Technology.org reports that the company has confidentially filed for a listing and recently led global large language model revenue at 31.4%, slightly ahead of OpenAI. The pricing of Fable 5 and Mythos 5 at USD 10 (approx. RM46) per million input tokens and USD 50 (approx. RM230) per million output tokens, twice Opus 4.8, underlines a premium positioning even as Fable 5 is framed as a broadly accessible model. At the same time, Anthropic is publicly calling for a shared “brake pedal” on frontier development and warning about possible recursive self-improvement. The dual launch of Claude Mythos 5 and Claude Fable 5 shows a strategy built on tiers: one path for cutting-edge research partners, another for mainstream adoption, both wrapped in assertive AI safety guardrails.
