Anthropic releases Fable 5.1 and Mythos 5.1, citing up to 45 percent savings on agentic workloads

Anthropic has introduced its latest models, Fable 5.1 and Mythos 5.1, which the company says respond to user complaints regarding pricing, data retention policies, and excessively strict safety filters.

Anthropic cuts pricing on Fable 5.1 by up to 45 percent for agentic tasks. The firm reports that the new version outperforms its predecessor while reducing typical expenses by around 25 percent and delivering savings of up to 45 percent on intricate agentic assignments, achieved through lower rates on previously processed and stored cached information.

Early users report gains in coding, speed and natural language. Every CEO Dan Shipper described it as “the strongest coding model we’ve used, but now it’s fast, token-efficient, and crucially actually speaks like a normal person.” Box CEO Aaron Levie reported that his firm’s agent powered by the update noticed subtleties and ambiguities overlooked by the prior release during identical evaluations.

Mythos 5.1 matches prior benchmarks at lower reasoning effort. Observer Lisan al Gaib noted that the variant running at low reasoning effort equals the benchmark results of its predecessor operating at maximum reasoning. The company is also introducing Mythos 5.1 inside Project Glasswing, where it will remain restricted to program participants.

Safeguards are adjusted and data retention improves for enterprise. According to Anthropic, Fable 5.1 includes “more precise safeguards” that are reportedly less prone than before to rejecting straightforward biology queries, whereas Mythos 5.1 carries forward identical biology limitations from the earlier edition. The firm further detailed advances on data retention, stating that its Enterprise Frontier Safeguards deliver “complete privacy” by keeping information on the client’s own cloud infrastructure rather than Anthropic’s servers; rollout is scheduled to begin later this fall.

Cybersecurity use expands with limits. The company indicated it is “now allowing Fable 5.1 to be used for identifying software vulnerabilities,” although certain related assignments—including penetration testing, exploit generation, and binary-based vulnerability scanning—will continue to route toward Opus models. Fable 5.1 can now be accessed across every platform.