跳到正文
RCreddit.com·
暂不在当前实时榜单

After Astra's stealth nerf last night, we really need benchmarks to do a re-bench 1 week after any model release. This is ridiculous.

AI 摘要

A user on reddit.com expressed frustration over a "stealth nerf" to the Astra model, claiming its capabilities significantly diminished overnight. They noted that Astra, which previously excelled at "1 shotting AAA game level models," now produces "completely terrible" results even after "reprompting on xhigh effort several times." The user criticized "frontier labs" for quietly reducing model capabilities to save compute, especially after many users purchased "Pro subscriptions" based on initial performance, suggesting the "FTC should get involved."

为什么是这条

This report highlights a user's direct experience of a model's performance degradation, unlike many discussions that focus on theoretical capabilities or benchmark scores.

时间与来源

时间显示为 UTC

显示时区:UTC

本地时区尚不可用,暂时显示 UTC。

收录当时偏移:UTC+02026年9月11日 04:00 UTC

收录
2026年9月11日 04:00
来源类型
开发者社区

讨论趋势

→ 平稳
最近 24 小时与此前 24 小时的快照均值对比 · 7 天曲线

百分比基于采集到的讨论信号,不代表新增评论数或独立参与人数。曲线仅用于同一话题在不同时段的比较。

正文

本站未收录正文。

前往源站阅读 →
来源·reddit.com