source: techcrunch ai: anthropic’s claude fable is a version of mythos the public can access today
level: technical
anthropic released claude fable 5, the first publicly available version of its mythos model. the model excels at software engineering, knowledge work, and vision tasks. it includes safety limits that block responses in areas like cybersecurity, biology, and chemistry, falling back to claude opus 4.8 instead. fable 5 is accessible through the claude api and enterprise plans, with subscription access rolling out in stages. after june 22, usage will require credits, though anthropic plans to restore it as a standard feature later.
the launch follows anthropic's warning about rapid ai progress and the risk of recursive self-improvement. the company stress-tested fable 5 with internal and external red-teaming, finding no universal jailbreaks in over 1,000 hours of testing. despite this, novel attacks remain possible. as a safety measure, anthropic now requires 30-day data retention on all traffic, even for enterprises with prior zero-retention agreements. the data will not be used for training, only for defending against attacks and reducing false positives.
early data shows at least 95% of fable sessions run entirely on the model's own responses. third-party testers report strong performance: hex noted a 90% score on a core analytics benchmark, base44 highlighted better one-shot app creation, and genspark said fable outperformed other models on ui design and game coding. pricing is $10 per million input tokens and $50 per million output tokens, double the cost of opus 4.8. anthropic expects high demand, and some enterprises see the cost as justified for autonomous operations.
why it matters: this release shows how frontier ai models are being deployed with built-in safety restrictions and mandatory monitoring, which could set new industry norms for data handling and access control.
source: techcrunch ai: anthropic’s claude fable is a version of mythos the public can access today