Updated
Updated · TechCrunch · Sep 1
Anthropic Launches Fable and Mythos 5.1, Cutting Token Costs and Expanding Zero-Data Retention
Updated
Updated · TechCrunch · Sep 1

Anthropic Launches Fable and Mythos 5.1, Cutting Token Costs and Expanding Zero-Data Retention

3 articles · Updated · TechCrunch · Sep 1

Summary

  • Anthropic on Tuesday released Fable 5.1 broadly via cloud platforms and its API, while keeping Mythos 5.1 limited to registered partners in cybersecurity and life sciences.
  • Fable 5.1 adds lower token costs, fewer false-positive safeguard blocks and Zero Data Retention support, letting clients run models on their own infrastructure without data outflows.
  • Enterprise Frontier Safeguards — a high-privacy service previously unavailable for Fable — will roll out in June, with Anthropic still monitoring misuse but letting clients control how that monitoring works.
  • Benchmark results set company records on Terminal-Bench 4.0 and Humanity’s Last Exam, and Anthropic said the models produced three scientific findings before release, including a GPU optimization and a Venus map.
  • The system card labels Mythos 5.1 low-risk on automated AI development, while noting a slight regression versus Opus 5 on overall misaligned behavior but improvements over Mythos 5 and Claude Sonnet 5.

Insights

If Mythos and Fable share the exact same AI brain, what dangerous capabilities are standard enterprise users secretly missing out on?
How can Anthropic guarantee to catch severe biological AI misuse if they never actually review the enterprise data?
With Humanity's Last Exam already being conquered, are we rapidly running out of ways to measure true artificial intelligence?