gemini-4VIDEO
Gemini 4 Argon 上手：展示页惊艳但上限偏低
实测者称 Argon 能一次生成漂亮展示页面，但任务一复杂就很快触顶，怀疑在 METR 类长程基准上表现不佳。
Binkg
@dummerspast3927 天前收录原推发布于 2026.10.01
Binkg @dummerspast39原推发布于 2026.10.01
First Gemini 4 Argon impressions: Creates beautiful showcases, but lopsided. It is surely trained on a lot of websites and one shot generations but it caps out quite fast. When giving it a too ambitions task it doesnt seem to be able to excede a certain threshold. you know this metr benchmark? feels like it wouldnt perform great there. might be the same for real work. CC: @Inmerson
x.com/dummerspast39/status/2105425737867985341采集于 2026.10.02 10:15