Everyday Style Imitation (Enron Email) - Authorship Verification: leaderboard
Tests whether a language model can imitate an ordinary person's writing style from a few of that person's real texts. Wang et al. (EMNLP 2025 Findings) give each model five random writing samples by one author plus a GPT-4.1 summary of a held-out text, and ask it to write that text in the author's voice at temperature 0. Enron: one-to-one internal emails by 150 employees, 3,884 test texts. Two authorship-verification models (Longformer-base-4096 and ModernBERT-base) fine-tuned on the corpus judge whether each generation and the author's real held-out text share an author; the score is the share judged same-author, averaged over both verifiers. On human test pairs the verifiers are 88.9% accurate. Verification accuracy (%), five-shot; higher is better.
Metric: AV Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o Mini | 96.64 | #598 |
| 2 | Gemini 2.0 Flash | 96.44 | #351 |
| 3 | DeepSeek V3 | 96.3 | #329 |
| 4 | Gemma 3 27B | 96.17 | #606 |
| 5 | GPT-4o | 96.15 | #357 |
| 6 | Llama 4 Maverick | 95.65 | #486 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=everyday-style-imitation-enron-email-authorship-verification · How It Works · Data refreshed daily, snapshot 2026-11-05.