Breaking

Anthropic releases Claude Fable 5.1 and gated Mythos 5.1, cutting typical costs about 25%

September 2, 2026 at 01:02 EDT

Foundation Models
Software Dev & Coding
AI Agents

Anthropic on September 1, 2026 released Claude Fable 5.1 and Claude Mythos 5.1, a Mythos-class pair that share the same underlying model and sit above the company's Haiku, Sonnet, and Opus lineup, with Fable 5.1 generally available and Mythos 5.1 limited to trusted-access programs.

Sep 1, 2026 · Anthropic

Cheaper cache, harder coding

Claude Fable 5.1 and Claude Mythos 5.1 share one set of weights. Fable is generally available; Mythos stays behind trusted-access gates.

Cache-read rate, drawn to scale

Fable 5.1 cache hits cost 0.025× the input price, versus 0.10× on Fable 5 — a 4× cut.

0.10×

Fable 5 cache rate

0.025×

Fable 5.1 cache rate

Takeaway: A cache hit now costs one-quarter of the Fable 5 rate.

Terminal-Bench-Science 0.1

Accuracy, stacked to scale. Fable 5.1 more than doubles Fable 5.

52.6%

Fable 5.1

29.0%

Opus 5

24.7%

Fable 5

22.4%

GPT-5.6 Sol

~25%

typical savings vs Fable 5

~45%

highly agentic savings, up to

75%

cheaper cache reads

$10/$50

input / output per million tokens

What testers like

Early-access teams call Fable 5.1 a stronger, faster coding model. Cognition said it would move Devin traffic from Opus 5 on launch day.

Where it still hurts

Developers report hitting session and weekly caps sooner. Docs warn of whole-file rewrites and one tool call per turn in agent loops.

Same weights, two doors

Fable 5.1

Generally available, with cyber and biology classifiers still on.

Mythos 5.1

Trusted-access only, with looser cyber and life-science rules.

Safer false positives

Biology flags on benign requests down 85%; cyber interventions down about 60%.

Continue reading

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