Anthropic has launched two new versions of its Claude model: Claude Fable 5 and Claude Mythos 5. Both are powered by the same underlying AI model but feature distinct safety guardrails, with Mythos 5 representing a new tier focused on reduced harmfulness.
Claude Fable 5 retains the standard safety measures, while Claude Mythos 5 applies stricter constraints, making it suitable for applications requiring higher reliability and lower risk of harmful outputs. This dual-release strategy allows developers to choose the balance between capability and safety.
The new Mythos-class tier is designed for high-stakes environments like healthcare, legal, and education, where even minor safety failures are unacceptable. Anthropic has not disclosed detailed benchmarks but states that Mythos 5 has been extensively tested for adversarial inputs and harmful content generation.
Both models are now available via Anthropic's API, with pricing reflecting the additional safety engineering in Mythos 5.