429/empty responses for gemini-3-flash-preview + function calling

Gemini API
1

The dashboard shows you’re nowhere near the limits:

  • RPM: 4 / 1,000 (0.4% used)

  • TPM: 3.08K / 1M (0.3% used)

  • RPD: 15 / 10,000 (0.15% used)

The Google AI studio dashboard says plenty of capacity is available, yet the API returns 429/empty responses specifically for function calling requests.

This is with thegemini-3-flash-preview + function calling. The model works for plain text but breaks when tools/function declarations are included.

2

I’ve been having this 429 error on all text models and I’m well below the quota. I think what is happening is that there is undocumented limit how many requests to the API (not the model) you can do in an undisclosed time frame.

My trouble is simply placing a new batch without getting close to the limits. If I send too many requests to place batches, I get error 429 which doesn’t explain what’s the actual problem is. So I suspect it’s about use of the API itself. Just poor implementation of the error.