🔶 请不要停用 Gemini 2.5 Flash

Nick_D 2026年7月10日 12:13 #1

大家好 -

首先，我想感谢 Gemini 团队提供了如此优秀的模型，这真的很有帮助。

我们有一些非常具体的工作流依赖 Gemini 2.5 Flash。我们的内部基准测试显示，Gemini 3 Flash 的表现不如前者（即使按照新的提示指南和其他变化尝试调整提示）。

我确信还有其他人有同样经历，而且没有简单的切换方案。

如果不停止使用 2.5 Flash，我们将非常感激。

是的，就连我们自己的基准测试也显示，在延迟和性能上最接近的模型——3.1 Flash Lite，也远远比不上 2.5 Flash。我们看到了思维泄露的问题。不得不说，2.5 Flash 是我们见过的全能表现最好的模型，在大多数任务中都非常好用。

我敢肯定，他们的大量流量和使用都来自这个模型。真心希望 Google 团队能更长久地保留该模型。

说得好。退役 2.5 Flash 将是巨大的损失。这是唯一在澳大利亚部署的低延迟模型。我能从 2.5 Flash 获得 300-400 毫秒的响应完成，使其适合语音代理。

3.5 Flash 需要 600-700 毫秒，而且没有澳大利亚部署，因此实际延迟接近 700-800 毫秒，完全破坏了语音代理的用例。

说真的，在这个地区部署的其他模型，在低延迟应用的质量上都无法接近 2.5 Flash。

我更担心从 Gemini 2.5 Flash 升到 3.5 Flash 的成本上升，后者大约贵 3 倍。我本以为 Flash 模型的意图是相对低延迟且比 Pro 更实惠，但新款 Flash 模型并未如此定价。话又说回来，廉价且充裕的 AI 时代或许要结束了……

Bcoun 2026年7月10日 22:11 #5

附议，我们的替代方案不是升级到 flash-3，而是找一个合适的开放权重模型

merc 2026年7月10日 22:26 #6

+1 我有些关键工作流，在同样的价格/智能水平上，没有其他模型能胜任！如果这个模型被停用，将是一记重击——我们可能会转向开源模型，但 2.5 Flash 的延迟很难被超越。

Nick_D July 10, 2026, 12:13am 1

Hi Everyone -

Firstly, I want to give my thanks to the Gemini team for providing access to such great models. It has been so helpful.

We have some very specific workflows that rely on Gemini 2.5 Flash. Our internal benchmarks show that Gemini 3 flash does not perform as well (even after attempting to tweak prompting following the new prompting guidelines and other changes).

I am sure there are others with the same experience as well, and there is no easy switch.

It would be extremely appreciated if 2.5 flash was not discontinued.

Yes, even our own benchmarks show that the closest model in latency + performance, which is 3.1 flash lite, doesn’t even come close to 2.5 Flash. Seeing issues with thoughts leaking out. I should say, 2.5 flash has been the best model we’ve seen for all round usage and works pretty well with most of tasks.

I am pretty sure, a huge chunk of their traffic and usage comes from this model. Would really appreciate it if the Google team can retain this model for much longer time.

Preach. Retiring 2.5 Flash will be such a massive loss. It’s the only low latency model that is deployed in Australia. I’m able to get 300-400ms completions from 2.5 Flash, making it suitable for voice agents.

3.5 Flash offers 600-700ms completions and doesn’t even get an Australian deployment, so the actual latency is closer to 700-800ms, completely breaking the voice agent use case.

There is honestly no other model deployed in this part of the world that comes close to the quality that 2.5 Flash offers for low latency applications.

I am more concerned about the cost step up from Gemini 2.5 Flash to 3.5 Flash, with the latter being roughly 3x more expensive. I thought the intention of the Flash models was to be relatively low-latency and more affordable compared to Pro, but the newer Flash models aren’t being priced as such. Then again, the era of cheap and plentiful AI might be coming to an end…

Bcoun July 10, 2026, 10:11pm 5

Seconded, the alternative for us is not upgrading to flash-3 but rather finding an appropriate open weight model

merc July 10, 2026, 10:26pm 6

+1 I have some critical workflows that no other model is good at for the same price/intelligence! It would be a huge hit to have this model discontinued - we’d likely switch to an open source model if this happened but the latency of 2.5 flash is something hard to beat.