Claude Fable 5.1: Same Model as Mythos, 75% Cheaper Cache Reads
Coffee Summary

Anthropic launched Claude Fable 5.1 (GA) and Claude Mythos 5.1 (trusted access) in September 2026.
Same underlying model; different safeguards. Model ID: `claude-fable-5-1`.
Base price unchanged vs Fable 5: $10 / $50 per 1M input/output; cache reads now $0.25/1M (75% less).
Anthropic estimates ~25% savings on typical workloads, up to ~45% on highly agentic ones.
EFS (customer-controlled privacy) rolls out in phases starting fall; ZDR available for eligible customers until EFS.
What happened
In its September 2026 post, Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1. They are the same underlying model with different safeguard levels.

**Fable 5.1** — generally available (`claude-fable-5-1` on the Claude API; also on AWS, Google Cloud, Microsoft Azure).
**Mythos 5.1** — trusted access only, with more permissive safeguards for vetted **cybersecurity** and **life sciences** work via CVP / LSVP programs.
Why it matters
Fable-class models are expensive when agent loops keep re-reading huge contexts. Anthropic’s cost move is concentrated on cache reads: same base IO rates as Fable 5, but cache hits get dramatically cheaper — which is where long agent sessions actually spend money.

What changed
Pricing
| Item | Fable 5.1 |

| — | — |

| Input | $10 / 1M tokens (same as Fable 5) |

| Output | $50 / 1M tokens (same as Fable 5) |

| Cache reads | $0.25 / 1M tokens (75% less than prior cache-read pricing) |

| Typical workload savings (Anthropic estimate) | ~25% vs Fable 5 |

| Highly agentic savings (Anthropic estimate) | up to ~45% |

Those savings percentages are Anthropic estimates from measured usage patterns, not a guaranteed invoice discount line-item.

Safeguards (high-level only)
Fable 5.1 can be used to **identify software vulnerabilities**, but **not to develop exploits** (Anthropic’s stated policy boundary).
Dual-use cyber tasks such as exploit generation remain redirected under Anthropic’s safeguards (details in their system card / post).
Life-sciences advanced capabilities for Mythos are gated through the **Life Sciences Verification Program (LSVP)**; cyber defender access via the **Cyber Verification Program (CVP)**.
This article intentionally stays high-level on biology/cyber safety — no dual-use detail.
Privacy path: EFS and ZDR
**Enterprise Frontier Safeguards (EFS):** customer-controlled cloud storage for data, aiming for privacy comparable to zero data retention while still enabling misuse detection; rolling out in phases beginning **later this fall**.
Until EFS is available, **eligible customers** can use Fable 5.1 with **zero data retention**.
Compliance extras
EU AI Act–related **watermarking** on outputs (models released after Aug 2, 2026 per Anthropic).
**Detection API** in **private preview** for eligible organizations.
Who should care
Teams on Fable 5 with cache-heavy agent/coding workloads — this is primarily a cost upgrade.
Enterprises waiting on ZDR/EFS clarity before moving traffic.
Security and life-sciences orgs that need Mythos-class access via CVP/LSVP rather than GA Fable.
Limitations
~25% / ~45% savings are **estimates**; your mix of cache reads vs new tokens will dominate.
Mythos is **not** a self-serve GA SKU.
EFS timing is phased (“later this fall”) — confirm eligibility rather than assuming day-one availability.
Benchmark tables on Anthropic’s post are extensive; treat them as vendor-reported unless you re-run them.
What to do next
1. Point staging traffic at `claude-fable-5-1` and compare cost per successful task vs Fable 5 (watch cache-read share).

2. If you need ZDR, confirm eligibility now and track EFS enrollment for fall.

3. If you need Mythos-only capabilities, apply to CVP or LSVP — do not expect Fable to mirror Mythos safeguards.

4. For EU compliance workflows, ask about Detection API private preview access if watermark verification is required.

AIImpish Take
Fable 5.1’s headline is not a base-price cut — it is cheaper memory. If your agents are cache-read heavy, Anthropic’s 75% cache-read reduction is the whole product. Mythos is the same brain with a different keycard. Price your real trace mix, lock ZDR/EFS status with sales, and ignore vanity benches until cost-per-solved-task moves in your logs.