Anthropic's new AI duo reduces costs and refines safety
Anthropic unveiled two new AI models, Fable 5.1 and Mythos 5.1, claiming lower costs and improved performance. The company says Fable 5.1 is up to 45% cheaper for complex agentic tasks and has more precise safeguards that are less likely to block basic biology questions. Data retention is addressed through Enterprise Frontier Safeguards that store data on customer cloud servers.
The pricing reduction stems largely from cheaper rates on cached data—information the model has already processed and stored—making repeated or complex agentic workloads significantly more economical. Early testers reported notable improvements: Box's CEO said its agent caught subtleties in data that the previous model missed, while Every's CEO praised both speed and more natural conversational output.
Mythos 5.1, the smaller companion model, achieves benchmark parity with its predecessor's maximum reasoning setting while using low reasoning, suggesting meaningful efficiency gains. Anthropic also clarified its cybersecurity stance, permitting Fable 5.1 for vulnerability identification while still routing penetration testing and exploit generation to its Opus line. Enterprise data controls arrive later this fall.
These models could reshape how businesses weigh AI adoption, as lower costs and improved data privacy may make enterprise deployment more attractive to cost-conscious organizations. The refined safeguards might reduce frustrating blocks on legitimate queries, but the continued restrictions on biology topics and selective cybersecurity permissions could raise questions about where Anthropic draws its safety boundaries. Smaller competitors may feel pressure to match these pricing and capability benchmarks.