Breaking

Anthropic Launches Claude Fable 5.1 and Mythos 5.1, Cutting Cache Prices and Lifting Coding Scores

September 1, 2026 at 17:48 EDT

Foundation Models
Software Dev & Coding
AI Agents

Anthropic on Tuesday released Claude Fable 5.1 and Claude Mythos 5.1, a Mythos-class pair it describes as its strongest models yet for coding and knowledge work, with cheaper cache reads and a large jump on agentic science tests. The two names are the same underlying model with different safeguards: Fable 5.1 is generally available, while Mythos 5.1 is limited to trusted-access programs for cybersecurity and the life sciences.

Sep 1, 2026 · Anthropic

Fable 5.1 retakes the public frontier

Anthropic released Claude Fable 5.1 and a restricted Mythos 5.1 twin: stronger coding and science scores, 75% cheaper cache reads, and the same $10/$50 token rates.

Agentic science pass rate

Terminal-Bench-Science 0.1

52.6%

Fable 5.1

29.0%

Opus 5

24.7%

Fable 5

22.4%

GPT-5.6 Sol

Fable 5.1 more than doubles its predecessor on this command-line science-agent test.

52.6%

Science bench (Fable 5.1)

75%

Cache-read price cut

~25%

Typical workload savings

$10/$50

Input / output per MTok

Magnitude

Cache-read price per million tokens

$1.00

Fable 5

$0.25

Fable 5.1

Cache hits now cost a quarter of Fable 5’s rate — Anthropic’s main lever for the estimated 25% typical and up to 45% agentic savings.

What testers liked

Early partners at Cognition, Jane Street, and Millennium reported stronger long-horizon coding, clearer multi-step traces, and cheaper Fable-class review after the cache cut.

The catch

Base rates stay twice Opus 5. Independent evals found max-effort runs can cost more than Fable 5, and remaining classifiers still reroute some cyber and biology queries.

Same weights, split access

Fable 5.1 is public. Mythos 5.1 is not.

Fable 5.1

General availability on API and major clouds. Tighter cyber and biology classifiers; can find vulns, not write exploits.

Mythos 5.1

Vetted U.S. cyber and life-sciences programs. More permissive safeguards; 60.9% on Terminal-Bench 4.0.

Continue reading

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