Skip to content
Back to model rankings

Benchmark sources

This page's content is in Chinese.

Find what each source measures, how it updates and whether it counts toward rankings.

24In this round's ranking72Reviewed sources and specialties
How ranks are calculated

综合评测与体验

17

跨能力的综合测试与真人盲选。

编程

15

从写代码到改仓库,看模型能不能把软件做出来。

推理

12

数学、逻辑与陌生规则,看模型能不能想明白新问题。

知识

5

事实问答与研究生级科学知识,看知识掌握与回答准确性。

专业办公

19

金融分析、法律咨询与银行业务,看专业任务能否完成。

视觉理解

2

理解图片的证据继续保留在综合榜;读懂图片与设计美观是不同能力。

中文与多语言

2

语言基础的补充证据,不把英文写作表现当作中文能力。

“Watching” means data, run conditions or usage limits are still being checked; these do not count toward overall or category rankings. Repeat fetches and display slices of one benchmark add no voting weight. Task overlap between different benchmarks is still reviewed continually.