# Fable 5.1 Cuts Cache Costs and Plans Customer-Held Monitors; Mythos Stays Gated

Times of AI Desk · 2026-09-01 · Models

[https://timesof.ai/2026/09/anthropic-claude-fable-5-1-mythos](https://timesof.ai/2026/09/anthropic-claude-fable-5-1-mythos)

> Anthropic launched Fable 5.1 with $0.25 cache reads and kept Mythos 5.1 on trusted access. Enterprise Frontier Safeguards were announced for a phased fall rollout, with interim zero data retention for eligible customers.

**Correction / update, October 14, 2026:** EFS was announced for a phased fall rollout, not opened generally on launch day. Removed an unsupported same-week comparison and added benchmark limitations.

Anthropic paired cheaper Fable 5.1 cache reads with plans for customer-controlled monitoring storage.

**Anthropic** (September 1) launched **Claude Fable 5.1** (generally available) and **Claude Mythos 5.1** — **same weights**, looser **cyber/bio** safeguards, **gated to trusted-access**. Input/output stays **$10 / $50** per million tokens; **cache reads** drop **75%** to **$0.25 / MTok**. Anthropic: typical workloads ~**25% cheaper**, highly agentic up to ~**45%**. Announced for a phased fall rollout: **Enterprise Frontier Safeguards (EFS)** can store monitoring data in the customer’s **AWS / Azure / GCP** account; Anthropic **does not charge** for EFS; **interim ZDR** on Fable 5/5.1 until fall rollout. Fable may now **discover vulns, not write exploits**. New API accounts **cannot edit prior context** while preserving the transcript of prior thinking (anti-distillation). Vendor: Terminal-Bench-Science 0.1 **52.6%** vs Fable 5 **24.7%** / Opus 5 **29.0%** / GPT‑5.6 Sol **22.4%**. Mythos 5.1 protein binders: ~**50%** hit rate vs typical **10–15%**. **Jane Street**: “solves more of our coding problems than Fable 5 or Opus 5.” **Cognition**: moving **Opus 5 Devin** traffic to Fable 5.1 on launch day. Mythos currently **US orgs**, expansion coordinated with USG.

## Claims vs checks

Lab benches (Terminal-Bench-Science, protein-binder hit rate) are **vendor**. Treat **52.6%** as **Anthropic-reported**. The company gives a 3.5–4.5 percentage-point standard error and notes that its setup differs from the public leaderboard; production safeguards also affect comparisons. Watch: **Mythos trusted-access** scale, and whether **Cognition’s Devin** flip shows up in usage.

## Sources

- [Anthropic: Introducing Claude Fable 5.1 and Claude Mythos 5.1 (September 1, 2026)](https://www.anthropic.com/claude-fable-and-mythos-5-1)
- [Anthropic: Developing Enterprise Frontier Safeguards (September 1, 2026)](https://www.anthropic.com/news/enterprise-frontier-safeguards)
- [Fable 5.1 / Mythos 5.1 system card](https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card)