Roadmap & 429 Fixes for Gemini 3 Pro Image

Google Cloud Build with AI Open Models
1

We are using gemini-3-pro-image-preview for a production project (target: 20 RPM). We frequently hit 429 Resource Exhausted errors, but Provisioned Throughput is currently too expensive for our scale.

  • What is the ETA for this model to move from Preview to GA (Stable)?

  • Besides Provisioned Throughput, what is the best way to stabilize 20 RPM ?

  • Will the stable version be launched in all regions simultaneously?

2

You can try the retry and exponential backoff logic for your API calls. If you’re using GenAI SDK, try:

client = genai.Client(
          vertexai=True, 
          project=PROJECT_ID, 
          location="global",
          http_options=types.HttpOptions(
              timeout=int(60*1000), # 60 seconds
              retry_options=types.HttpRetryOptions(
                  attempts=3, 
                  initial_delay=1.0, 
                  http_status_codes=[408, 429, 500, 502, 503, 504]
              )
          )
        )

Yes, Provisioned Throughput is the best option.

Stable versions typically come with regional support, and in a gradual expansion.

1 Like