Home /Benchmarks, evals and calibration /jev-report

jev-report

Independent Chinese research report: 52 pages, 50 reproducible tests, 143 traceable data rows.

From the readme

发明 RLHF 的人,这次做了个不会说话的模型 两千多万人围观的 Jev,我用中文测了 50 条。 52 页 PDF,10 张图,143 条能回溯到出处的数字。免费,随便转。 下载 PDF · 章节源文件 · 数据表 · 50 条中文实测 它是什么 Jev 是 TypeSafe 在 2026 年 9 月 15 日发布的模型。创始人 Diogo Almeida 是 RLHF 的共同发明人之一。 这个模型不生成文字。你给它一段材料和一道题(选哪个、打几分、是不是),它返回选项和概率。输入价每百万 token 0.042 美元,输出免费。主页写它比前沿模型快 193.6 倍、便宜 444.6 倍。 Hacker News 上的发布帖拿了 1,831 分、480 条评论,一半在吵。 …

Details

Section
Benchmarks, evals and calibration
owner
HackSing
stars
0
forks
0
pushed
2026-09-17
Language
Python
License
MIT

What it is

Form
Dataset or benchmark
Audience
Researchers
Maturity
Docs
reports measured numbers

Useful for

Best intent matches