Felda AI Builds Fine-Grained Writing Benchmark Spanning Rhythm and Naming Habits

On September 7, the Felda AI team (author @almmaasoglu) shared a series of posts on X about their in-house writing evaluation benchmark and its initial results: instead of scoring overall writing style, the evaluation drills into concrete features of a model's writing, including naming preferences, opening word choices, punctuation habits, and the "sentence rhythm" of alternating short and long sentences within a text, with early charts for the sentence-rhythm dimension.

Confirmed

The team ran the same sentence-rhythm analysis on five models, comparing how much each mixes short and long sentences within an article.
Their conclusion: average sentence length carries limited information; the real differences between models show up in the diversity of sentence rhythm.
An interesting finding: OpenAI and Anthropic models both favor the same name; each scatter plot covers 20 outputs, with lines connecting the same prompt group across models to visualize naming preference distributions.

Not yet confirmed

A fuller official benchmark report and more findings have not been published yet; the author says they are coming.

Why it matters

Mainstream writing evaluations mostly focus on macro style or win rates; this benchmark pushes granularity down to quantifiable features like punctuation, naming and sentence rhythm, offering a new lens for distinguishing models' writing personalities and highlighting the limits of common metrics like "average sentence length."

2026-09-07 ~ 2026-09-07 · 5 related posts

Primary sources

Five-model writing style comparison shows average sentence length misses the mark — almmaasoglu · 2026-09-07 00:36 (UTC+08:00)
OpenAI and Anthropic models share a favorite name, writing benchmark finds — almmaasoglu · 2026-09-07 00:37 (UTC+08:00)
Felda AI builds writing benchmark analyzing names, punctuation and sentence rhythm — almmaasoglu · 2026-09-07 00:35 (UTC+08:00)

AI startup Felda dissects writing quality down to sentence rhythm and punctuation in its new benchmark — almmaasoglu · 2026-09-07
A custom AI writing benchmark examines sentence rhythm across five models — almmaasoglu · 2026-09-07
[source] Felda AI builds writing benchmark analyzing names, punctuation and sentence rhythm — almmaasoglu · 2026-09-07
[source] Five-model writing style comparison shows average sentence length misses the mark — almmaasoglu · 2026-09-07
[source] OpenAI and Anthropic models share a favorite name, writing benchmark finds — almmaasoglu · 2026-09-07