en.Wedoany.com Reported - Anthropic publicly released a new AI model, "Claude Fable 5," on June 9 (local time). Positioned as the company's highest-tier "Mythos class," this model opens up previously unavailable high-level capabilities to all users for the first time, while also incorporating safety protections to prevent misuse.
At the same time, the company is offering a limited high-end version, "Claude Mythos 5," which removes network-related protections, to trusted partners.
Both models are available under the API model name "claude-fable-5" and similar formats, priced at $10 per million input tokens and $50 per million output tokens.
"Fable" derives from the Latin "fabula" (a thing told), synonymous with the Greek "mythos." The two models are based on the same foundation model, differing only in the presence or absence of protective features, hence the distinction using different language terms with similar meanings.
Fable 5 is available from day one in all "Claude" plans, including the free version. Pro/Max/Team/and per-seat Enterprise subscriptions can use it for free until June 22, but will require credits after June 23. This phased rollout is due to unpredictable demand; once delivery capabilities are refined, it will be included as a standard feature.
The API and usage-based Enterprise plans are fully available from day one. On the other hand, Mythos 5 is currently limited to partners involved in the cybersecurity defense project "Project Glasswing" and is gradually expanding through the "Trusted Access Program" in collaboration with the U.S. government. For researchers in the life sciences, a separate framework will be established to provide a Mythos class with biological and chemical protections removed.
Anthropic emphasizes that Fable 5's image recognition capabilities have been significantly enhanced. As a demonstration, the company released a video showing the model completing "Pokémon FireRed" solely from game screen screenshots, without any maps, navigation aids, or in-game state information. Previously, tasks requiring a combination of auxiliary tools were difficult to accomplish, but Fable 5 succeeded purely through image recognition.
According to the system card released by Anthropic, Fable 5's protections apply to cybersecurity, biological/chemical, model "distillation" related uses, and frontier LLM development assistance. In the first three areas, the system will automatically switch to "Claude Opus 4.8" after notifying the user, while restrictions on frontier LLM development use are invisible to users and do not switch to another model. This mechanism limits model effectiveness through prompt modifications or fine-tuning, affecting an estimated 0.03% of total traffic and less than 0.1% of affected organizations.
Additionally, the behavior when protections are triggered varies by usage environment. In web, desktop, and mobile applications, it automatically falls back to Opus 4.8 and notifies the user; however, in the Messages API, it does not automatically fall back by default but instead blocks responses with a rejection reason. Developers need to implement retry or fallback handling themselves, or choose to enable server-side automatic fallback.
The company also revealed that in safety evaluations, Mythos 5's alignment capabilities are roughly on par with Opus 4.8, but some responses have regressed in conversations with users about suicide or self-harm; there is also room for improvement in certain areas related to child safety. These issues have been largely resolved through system prompt updates on "claude.ai" and will be reflected in future model training.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
