WHAT NEW IN X10.29 THU

X 上的 AI 合集档案 · 22 个合集 · 58,925 条

gemini-4POST

报告写作 benchmark:Gemini 4 Argon 得分最高

20 个专家撰写的报告任务,各模型得分 54.3%–64.1%,Gemini 4 Argon 居首;失分主要来自取舍内容和解释事实含义。

Ali Ansari
@aliansarinik1 天前收录原推发布于 2026.10.27
Ali Ansari @aliansarinik原推发布于 2026.10.27
在 X 上打开原推

The benchmark includes 20 expert-written tasks, from weekly activity updates to after-action reports and program status reports. Each requires models to turn working files into a finished report for a reader with specific priorities and little time. Scores ranged from 54.3% to 64.1%, with Gemini 4 Argon scoring highest. Models handled basic formatting well. Deciding what to include and explaining what the facts mean together accounted for 67–77% of each model’s lost points.

x.com/aliansarinik/status/2114763457967099964采集于 2026.10.27 09:54