Breaking

Anthropic’s Claude Fable 5.1 Draws Early Praise for Website Output as Token Costs Fall Versus Fable 5

September 2, 2026 at 03:46 EDT

Foundation Models
Software Dev & Coding

Anthropic’s newly released Claude Fable 5.1 is already being used to generate websites, and early builders say it is producing stronger front-end output than other AI models they have tried—while running cheaper than Fable 5 on typical token-billed jobs.

Sep 1, 2026 · Anthropic

Claude Fable 5.1: cheaper Fable, sharper sites

The generally available Fable-class model is drawing early praise for website and interface output, while typical token-billed jobs run about 25% less than Fable 5.

~25%

lower typical cost vs Fable 5

~45%

lower cost on agentic work

75%

cheaper cache reads ($0.25/MTok)

+17.2%

iterative code fixing at high effort

Terminal-Bench-Science 0.1

Accuracy, drawn to scale. Fable 5.1 more than doubles Fable 5 on this agentic science eval.

24.7%

Fable 5

52.6%

Fable 5.1

Cache-read price, to scale

A 75% cut, from $1.00 to $0.25 per million tokens, is why long website and coding loops get cheaper even though base input and output rates did not change.

$1.00

Fable 5 cache read

$0.25

Fable 5.1 cache read

Where it is winning

Cleaner website and UI output, more human copy, and a real edge on live-app fixes. Self-checks in the browser before a task is marked done.

Where it still lags

From-scratch (0-to-1) scores slipped slightly in Lovable tests. Code review was 48.7% slower. Base rates stay at $10 / $50 per million tokens.

How a site get built

Fable 5.1 is strongest when the page already exists and must keep working.

Brief the page

→

Generate HTML/CSS/JS

→

Open it in the browser

→

Iterate the live app

Continue reading

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