问 HN：只有我一个人觉得 Gemini 在过去三周里变烂了吗？
2 分•作者: EvanMcCormick•大约 1 个月前
作为 Gemini Pro 计划的 9 个月用户，我注意到从大约五月初开始，我从 Gemini 获得的速率限制大幅增加。
这似乎与更新的 UI 和 Gemini 3.5 预览模型的发布时间相吻合。
今年大部分时间里，Gemini 一直是我快速回答简单问题的首选模型。谷歌搜索插件让提问变得非常容易，而且在每月 20 美元的计划中，它似乎拥有无尽的研究和复杂答案的能力。与此同时，我很少使用 Claude，并将我的 Pro 计划的大部分用于 Claude Code 代币。现在我发现 Gemini 对我的速率限制比 Claude 更严格。计算资源是否正在幕后从 Gemini 转移？
也许 LLM 的计算资源最终正在远离补贴的直接消费者使用。尽管如此，我之前认为 Gemini 会是这种用例中最后的幸存者，因为谷歌似乎有最多的钱可以烧。
查看原文
As someone who's been using the Gemini Pro plan for the past 9 months, I noticed a massive jump in the amount of rate-limiting I'm getting from Gemini since around the beginning of May.<p>It seems to coincide with the updated UI and the release of the Gemini 3.5 Preview model.<p>For the better part of this year, Gemini was my go-to model for answering simple questions quickly. The google search plug-in made it super easy to ask, and it seemed to have endless capacity for research and complex answers on the $20/month plan. Meanwhile, I used Claude very sparingly, and saved most of my Pro plan for Claude Code tokens. Now I find that I'm rate-limited by Gemini harder than Claude. Is compute being shifted away from Gemini behind the scenes?<p>Maybe LLM compute is finally moving away from subsidized direct consumer use. Still, I had previously thought that Gemini would be the last man standing in that use case, as Google seemed to have the most money to burn.