Anthropic released Claude Fable 5 on Tuesday afternoon. This model is the first publicly available model in Anthropic's Mythos model family.
Claude Fable 5 utilizes the same technology tier as the Mythos models and includes operational safeguards. The model contains classifiers that restrict automated responses to prompts related to cybersecurity, biology, and chemistry. When the system restricts a query, it automatically diverts the prompt to Claude Opus 4.8. Early usage data indicates that at least 95 percent of Claude Fable 5 sessions run entirely on the new model without triggering a fallback to Claude Opus 4.8.
Anthropic said in a reply to a news organization, "Fable 5's capabilities in areas like cybersecurity, biology and chemistry are advanced enough that we're taking a deliberately conservative approach for these topics at launch." An internal bug bounty program testing Claude Fable 5's classifiers logged over 1,000 hours without discovering universal jailbreaks. External security organizations also conducted red-teaming on Claude Fable 5 and did not identify universal jailbreaks.
Dianne Penn, Anthropic's head of product management for research, said, "For us, it's really around what we call 'race to the top,' being able to provide this technology in a valuable fashion, and at the same time providing the right safety guardrails so that it can do asymmetrically more benefits than harm." She added, "What we wanted to do was to be very intentional about building new types of classifiers and new types of safety guardrails in place for this launch."
Anthropic will provide Claude Fable 5 to users on Pro, Max, Team, and seat-based Enterprise plans at no additional cost until June 22. After this date, accessing Claude Fable 5 will require purchasing separate computing credits. The model is priced at $10 per million input tokens and $50 per million output tokens, which is double the rate of Claude Opus 4.8. Anthropic will require a 30-day data retention policy for all traffic processed by Claude Fable 5 and Mythos 5. The company stated the retained data will be used exclusively to identify attacks and reduce false positives, not for model training.
Anthropic stated in a blog post that "Fable'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." Anthropic also wrote, "The same queries that are beneficial in the hands of cybersecurity professionals and biology researchers could be dangerous if available to malicious actors." The company stated, "Mythos 5 is the company's first model to consistently produce novel, compelling scientific hypotheses." Claude Fable 5 scored more than 10 percent higher than Claude Opus 4.8 on certain performance benchmarks. Hex said in a statement, "On the hardest questions, it shows strong judgement and attention to nuance."
No independent assessment was available for this report.