Claude Fable 5.1: Cheaper, Calmer, Watermarked

The public Mythos-class model gets cheaper, raises fewer false alarms, and starts watermarking its text.

Anthropic releases Claude Fable 5.1, a point release that improves on Fable 5 at higher effort settings, cuts typical costs by about a quarter, sharply reduces cybersecurity false positives, and adds invisible watermarking to generated text.

Key Numbers

Typical cost vs Fable 5

About 25% lower

Cyber false positives

About 60% fewer

Verified Facts

• Fable 5.1 matched or beat Fable 5 at low and medium effort settings and performed much better at higher effort.
• It cost about 25% less than Fable 5 for typical workloads, and up to 45% less for highly agentic work, through cheaper cache reads; base prices stayed the same.
• Claude Code users saw around 60% fewer cybersecurity false positives, the safeguard refusals that had frustrated developers using Fable 5.
• The release added invisible watermarking of Claude-generated text. As with Fable 5, a less restricted twin, Mythos 5.1, went only to US entities in trusted access programs.

Frequently Asked Questions

Sources & Further Reading

• Anthropic Launches Claude Fable 5.1 With Lower Costs and Fewer False Positives (MacRumors, Sept 1, 2026)

Cite This Page