Sudden 429 errors

Gemini API
1

I’ve been using the gemini-2.0-pro-exp API for several months with no problems.

It’s the “Gemini AI” module of Make, so I’m not sure which endpoint exactly, but it involves Role User and Parts with Message Type “Text” and a "Text’ value, like a chat completion.

Today, I suddenly started getting:

[429] You exceeded your current quota. Please migrate to Gemini 2.5 Pro Preview (models/gemini-2.5-pro-preview-03-25) for higher quota limits. For more information on this error, head to: https://ai.google.dev/gemini-api/docs/rate-limits.

Even if I do switch it to gemini-2.5-pro-preview-03-25 as advised, I continue to get the exact same 429 error message.

In Google Cloud Console, where should I check?
In https://console.cloud.google.com/iam-admin/quotas, for project "Generative Language Client, I see a range of duplicate entries for Generative Language API and Gemini for Google Cloud API across different regions, but nothing has high usage; all tiny.

ChatGPT tells me:

  1. “Google has begun strictly enforcing usage limits on older Gemini models, such as gemini-2.0-pro. Even minimal usage can now trigger quota errors.”
  2. "Some users report that requests made to newer models like gemini-2.5-pro-exp-03-25 are being incorrectly attributed to older models in the usage dashboard, leading to premature quota exhaustion. "
  3. “The token-per-minute (TPM) limits for certain models have been lowered. For instance, gemini-2.5-pro-exp-03-25 previously allowed 1 million TPM but has been reduced to 250k TPM, causing 429 errors for larger payloads.”

What’s going on, please?

2

To your very disappointment, google effectively banned 0325 access yesterday when they announced that no more free tier access. As 0325 exp is only a free model, so it’s now inaccessible.

You can contribute to the ‘bring 0325 back’ thread to raise this issue. Google has been silent

3

Hi @Fr_L, some good news: I thought the same as you, but after some investigation, I found a way around this to access the 03-25 model checkpoint again. Please see my findings here:

1 Like