Breaking

Anthropic launches Claude Fable 5.1 and Mythos 5.1, cutting typical token costs by about 25%

September 1, 2026 at 14:41 EDT

Foundation Models
Software Dev & Coding
AI Agents

Anthropic on Tuesday released Claude Fable 5.1 and Claude Mythos 5.1, two Mythos-class systems it describes as its most advanced models for coding and knowledge work, with cheaper long-running agent loops and tighter safeguards than the June Fable 5 / Mythos 5 pair. The company says the two names are the same underlying model with different safety layers: Fable 5.1 is generally available, while Mythos 5.1 remains limited to trusted-access programs for cybersecurity and the life sciences.

September 1, 2026 · Anthropic

Fable 5.1: same sticker, cheaper loops

Anthropic released Claude Fable 5.1 for general use and Claude Mythos 5.1 for vetted labs. Same weights, 75% cheaper cache reads, and a science score that more than doubled.

25%

Estimated typical cost cut vs Fable 5

75%

Cheaper cache reads ($0.25 / MTok)

52.6%

Terminal-Bench-Science 0.1 (was 24.7%)

$10/$50

Unchanged input / output list price

Agentic science, drawn to scale

Terminal-Bench-Science 0.1 accuracy. Each block is 5 points; Fable 5.1 is more than twice Fable 5.

52.6%

Fable 5.1

24.7%

Fable 5

29.0%

Opus 5

22.4%

GPT-5.6 Sol

Cache reads now cost one-quarter as much

Fable 5 billed cache hits at $1.00 / MTok. Fable 5.1 bills $0.25. Agentic jobs that reread a repo pay the difference.

$1.00

Fable 5

$0.25

Fable 5.1

Takeaway: 4× cheaper cache reads; up to ~45% off highly agentic workloads.

What testers like

Cognition is moving Devin’s Opus 5 traffic to Fable 5.1 on day one. Jane Street reported more coding wins and readable long runs; Millennium said it traced a years-old crash no prior model found.

What still bites

CodeRabbit saw review latency jump ~49% and still advises selective use. Safeguard hits still zero some OSWorld and automation tasks. Mythos 5.1 stays U.S.-vetted only.

Same model · two gates

Fable 5.1 is public. Mythos 5.1 is trusted-access only.

Fable 5.1

General availability. Can find source-code vulns, not write exploits. Dual-use bio still routes to Opus.

Mythos 5.1

Project Glasswing and verification programs. Looser cyber and biology limits. 30-day retention by default.

Still blocked

Penetration testing, exploit generation, and binary-based vulnerability scanning on the public Fable path.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.