Anthropic launches Claude Fable 5, a “safe” Mythos-class model with safeguards attached

Anthropic has released Claude Fable 5, a Mythos-class artificial intelligence model, now available for general use worldwide. Users accessing Fable 5 benefit from its expanded capabilities, although the model employs stricter safeguards to prioritize safety. When certain topics are queried, users may receive responses from an alternative model, Claude Opus 4.8, due to these cautionary measures. Anthropic notes that while these protections are robust enough for general availability, they may occasionally block benign requests, and the company intends to minimize such false positives through ongoing updates.

Following this launch, Anthropic is introducing Claude Mythos 5, which uses the same underlying architecture as Fable 5 but with less restrictive safeguards. Initially, Mythos 5 will be provided to Glasswing partners with relaxed cyber-related controls and, in the near future, to select biology researchers with fewer restrictions on biology and chemistry topics. This exclusive access is limited until Anthropic expands its trusted access program.

Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens, representing a significant drop from previous models. Additionally, these new iterations operate autonomously for longer durations without human intervention compared to earlier Claude releases.