It’s here. Anthropic’s Mythos model, supposedly too dangerous for public release in April, is now available to wreak havoc or tackle other tasks for a hefty price and with some new guardrails in place. Just make sure you don’t mind having Anthropic keep some of your data for a while. The AI biz on Tuesday announced public availability of Claude Fable 5, a Mythos-class model, and private availability of Claude Mythos 5 for Glasswing partners. Both are distinct from Mythos Preview, the model family’s elder sibling. Mythos-class models are said to be a tier above Claude Opus in terms of benchmark performance. “Fable 5’s capabilities exceed those of any model we’ve ever made generally available,” the company said. “It is state-of-the-art on nearly all tested benchmarks of AI capability, showing exceptional performance in software engineering, knowledge work, vision, scientific research, and many other areas.” As if to suggest that the current technocracy still has room for humanities scholars, Anthropic notes that the name Fable comes from the Latin fabula, “that which is told,” similar to the Greek mythos. “The safeguards are what distinguish the two models (Fable and Mythos) and are why we’ve given them different names,” the company said. Anthropic considers Fable capable of causing serious damage without safeguards. So part of its protocol is to failover to Opus 4.8 for certain types of queries – prompts related to cybersecurity, biology and chemistry, or distillation. Fable 5 ships with a new set of classifiers, which are separate AI models that look for misuse. The AI biz is also instituting a new data retention policy – it’s retaining log data to have a record in the event of misuse. This applies specifically to organizations with zero data retention policies – which now technically are not zero data retention. “Prompts submitted to, and outputs generated by, Mythos-class models are retained for 30 days for trust and safety purposes, on every platform where these models are offered,” the company explains on its updated support page. Consumer plans (Claude Free, Pro, and Max) are unaffected – because Anthropic already has a data retention policy in place for those tiers. “This change only applies to organizations that have set up workspaces with zero data retention (ZDR) in Claude Console, use Claude Code with ZDR in Claude Enterprise, or access Claude through AWS Bedrock, Google Cloud Agent Platform, or Microsoft Foundry with ZDR,” Anthropic’s documentation explains. Temporarily retaining prompts and outputs is necessary to prevent misuse, Anthropic insists, adding that it won’t use this data for training new models. In terms of benchmarks, the two models significantly outperform Anthropic’s own Opus 4.8, Anthropic’s Codex 5.5, and Google’s Gemini 3.1 Pro on the chosen test suite. Not listed is Google’s more recent Gemini 3.5 Flash [PDF]. Fable 5 is said to excel at software engineering. Anthropic claims that Stripe used the model to migrate its 50-million line Ruby codebase in just one day. That project, Anthropic claims, would have taken a Stripe team two months without AI help. Fable 5 is also supposedly more token efficient than than its predecessors, which might somewhat mitigate the higher token cost. Fable 5 and Mythos 5 both charge $10 per million input tokens and $50 per million output tokens, which is less than half the price of Claude Mythos Preview ($25/m input, $125/m output). Mythos 5 delivers the strongest results yet for resisting prompt injection attacks, which saw a success rate of 4.8 percent over 100 attempts, per the model safety card [PDF]. That’s comparable to Claude Opus 4.8. As of today, until June 22, Fable 5 is available through Pro, Max, Team, and seat-based Enterprise plans at no extra cost. After that, Fable 5 will no longer be available through those plans and will require usage credits. At some point thereafter, “if capacity allows,” Anthropic intends to restore Fable 5 access in subscription plans. ®