# Claude Fable 5.2 Leads AA-3D Leaderboard With 1,271 Elo

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/53486
Published: 2026-11-11T02:22:09.000Z
Updated: 2026-11-11T02:22:09.000Z
Section: Technology

> AA-3D uses blind comparisons across 55 interactive scene-building tasks, so its Elo ratings measure performance on that benchmark rather than overall model quality.

A Nov. 10 check of the AA-3D leaderboard placed Anthropic’s Claude Fable 5.2 first at 1,271 Elo, giving the model a lead of 67 Elo points in an interactive 3D scene-generation benchmark.

Claude Sonnet 5.5 ranked second at 1,204, with a 95% confidence interval of 1,176 to 1,232. Claude Opus 5.5 followed at 1,202, with an interval of 1,174 to 1,231. Fable’s 95% interval ran from 1,246 to 1,297, and its score was 69 Elo points above Opus.

AA-3D v1.0 contains 55 tasks. In each task, a model uses three.js to turn a written prompt into an interactive web page in an offline sandbox. A recruited panel compares the results in blind pairwise matchups, without being told which model made each scene.

The resulting Elo rating reflects relative performance across those comparisons. The leaderboard therefore measures results within this scene-building task set; it does not rank models across every kind of work.

The benchmark’s Oct. 12 launch results put Sonnet 5.5 first at 1,194 Elo, statistically tied with three other Anthropic and OpenAI models. The Nov. 10 leaderboard check showed Fable 5.2 at the top, but the listing did not display when its evaluations were run or when the scores were updated.

Anthropic announced Fable 5.2 on Nov. 9 and listed it as available across its platforms. The company set API prices at $8 per million input tokens and $40 per million output tokens, describing both rates as lower than Fable 5.1 without specifying the size of the reductions.

Separately from the AA-3D result, Anthropic said Fable 5.2 targets demanding technical work and can generate outputs up to 30% faster than Fable 5.1. That speed figure is a product claim, not a result measured by the benchmark.

Sam Zhang, a creative technologist at Unity, described the model as “very strong on the harder game-dev tasks, where it keeps iterating and improving instead of getting stuck.”