AI newsSeptember 1, 2026

September 2026

Fable 5.1 and Mythos 5.1: same list price, a quarter of the cost on cache reads, far fewer false positives.

August 2026

Anthropic introduces Fable 5.1 and Mythos 5.1. The interesting part is not the benchmark line but the price of a cache read and the number of safeguard interventions per session. Anyone running long agent sessions feels both immediately. Alongside it, Enterprise Frontier Safeguards moves misuse detection into the customer's own cloud account. As always, every link goes to the primary source.

What moved

Releaseanthropic.comSep 1

Claude Fable 5.1 and Mythos 5.1 (opens in a new tab)

One model in two grades: Fable 5.1 generally available, Mythos 5.1 with more permissive safeguards and only for vetted cybersecurity and life-sciences organizations. Input and output stay at $10 and $50 per million tokens, cache reads drop from $1 to $0.25. Anthropic puts Terminal-Bench 4.0 at 55.8 percent against 42.0 for Fable 5 and Terminal-Bench-Science at 52.6 against 24.7, and counts around 60 percent fewer safeguard interventions per Claude Code session. Available as claude-fable-5-1 on the API and through AWS, Google Cloud, and Microsoft Foundry.

Same list price, a quarter of the cost on cache reads, and fewer stops mid-session.

What you can run

Toolinganthropic.comSep 1

Enterprise Frontier Safeguards: misuse detection in your own bucket (opens in a new tab)

Monitoring data sits in the customer's own cloud account, encrypted under the customer's keys, and alerts go straight to the customer's security team. Anthropic reads none of it. Opt-in, no surcharge, no effect on model behavior or rate limits, supported across Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Google's Agent Platform, and Microsoft Foundry. Anthropic rolls it out in phases from late fall. Until then, eligible customers get zero data retention on Fable 5 and 5.1.

The usual trade between monitoring and data control does not apply here.