The GPT 6 Astra Downgrade Was Real. OpenAI Acknowledged And Fixed It (Partially)
Reddit上的一篇帖子声称GPT 6 Astra的降级是真实的，OpenAI已部分承认并修复。作者指出，这似乎是一种模式：发布模型，等待基准测试结果，然后降级以最大化利润，并使下一个模型看起来更好。此外，由于Astra的热度导致计算资源短缺，OpenAI已暂停200美元的订阅，这暗示了可能存在不那么明显的限速措施。
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。
发布当时偏移：UTC+02026年9月12日 13:56 UTC
收录当时偏移：UTC+02026年9月12日 19:01 UTC
- 发布
- 2026年9月12日 13:56
- 收录
- 2026年9月12日 19:01
- 来源类型
- 开发者社区
- 档位
- 社区
- 信源状态
- 正常
档位是按信源手工设定的编辑判断，不是逐条打分。
注册后，可在设置中选择默认翻译语言。
a lot of people are sick of downgrade posts, and i get it. many of these posts are based on vibes and placebo and there's multiple of them a day.
but this is an obvious case of an actual downgrade. downgrades are a fact and happen more often than most of us notice. they are anti consumer and it's important that we, as customers, speak up about them.
it's always the same pattern: release a model, wait until the benchmarks are in (and the hype-subscribers gave you their money) and then downgrade it, to maximize profits and make your next model look better.
it's like selling a cleaning service and, after a month, starting to only clean half the apartment, for the same price. stuff like that should be illegal, if it isn't already. (but i assume that they put some kind of little trick in the TOS to bypass that.)
maybe benchmarks should start to retest current models regularly. ai labs would probably never dare to downgrade their models if that resulted in suddenly falling off the leaderboards.
openai already paused the 200$ subscription because they're running out of compute, due to the astra hype, so it's probably not unreasonable to assume that they could also be rate limiting us in less obvious ways.