Anthropic releases updated models with reduced costs and adjusted safeguards
Anthropic has rolled out updated versions of its advanced AI models, featuring lower token costs, improved reasoning benchmarks, and modified safety restrictions. The release balances high-performance capabilities with expanded privacy options for enterprise customers.
Published by Jin · 2 min read · 2 SEPT 2026
- 30-day
| Metric | OpenAI | Anthropic |
|---|---|---|
| Data retention policy | Zero data retention with Private Safety Processing | 30-day data retention |
| Models mentioned | — | Fable 5 and Mythos 5 |
On Tuesday, Anthropic introduced Fable and Mythos 5.1, marking the latest iteration of its advanced artificial intelligence models. Alongside performance improvements, the release addresses key enterprise concerns by lowering token costs and refining how safety safeguards handle benign prompts.
Model availability and access
Access to the new models remains tiered based on their intended use cases. Mythos 5.1 is restricted to registered partners conducting research in cybersecurity or life sciences. Meanwhile, Fable 5.1 is widely available starting today through major cloud platforms and the Anthropic API.
Privacy and data retention
Anthropic is also expanding its data privacy framework. A high-privacy service called Enterprise Frontier Safeguards is scheduled to roll out in the fall. This offering allows clients to run models with zero data retention and manage their own infrastructure without standard data outflows, while still maintaining oversight tools to prevent misuse.
Benchmark performance
Source — Original announcement ↗
Worth a read?
Comments · 0