GenAI service_tier - is it even working??

[danielLublinsky]

Hey,

Gemini has Priority inference - https://ai.google.dev/gemini-api/docs/priority-inference#how-to-use
And we recently allowed for priority for all of our AI uses
Sadly I cant get it to preform, we still get:

503 UNAVAILABLE. {'error': {'code': 503, 'message': 'This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later.', 'status': 'UNAVAILABLE'}}

The service_tier flag is set and the response acknowledges the setting - we are running Tier 2 on the API key:

Priority inference is available to Tier 2 & Tier 3 users across the GenerateContent API and Interactions API endpoints.

And still nothing,
I know that its not 100% grantee to give a response at peak-times but for the few weeks its out its completely useless and never helps avoid the 503 or speed response times like they promise

Am I possibly missing something?
did anyone else use this and actually get something out of it?

Environment details

• Programming language: python 3.14.4

• Package version: 2.1.0

Thanks!

[Venkaiahbabuneelam]

Hi @danielLublinsky, Thanks for reaching out us!

Could you please let us know which models are you using, so that we will try to reproduce from our side also.

Thanks

[danielLublinsky]

Hey,
We run gemini-flash-3.0 and gemini-pro-3.1

[danielLublinsky]

Any updates?
What can we do?
@Venkaiahbabuneelam

[Venkaiahbabuneelam]

Hi @danielLublinsky, Apologies for my late response.
We didn't encounter any error while switching the model. If the issue still persists, could you please share a minimal reproducible code snippet?
Thanks

[danielLublinsky]

@Venkaiahbabuneelam
Yes, we had this issue last week:
self.gemini_service_tier = types.ServiceTier.STANDARD or PRIORITY...

Api call:

       resp = self.chat.send_message(
            user_text,
            config=types.GenerateContentConfig(
            service_tier=self.gemini_service_tier,
            cached_content=self.cache_name,
            stop_sequences=["[DONE]"],
            temperature=temperature,
            thinking_config=types.ThinkingConfig(
                thinking_level=thinking_level
            )
        ))

Still hitting the 503s when the model is under load,
the resp dose include a flag service_tier acknowledgment standard or priority

[Venkaiahbabuneelam]

Hi @danielLublinsky, Apologies for the inconvenience. We encountered a 503 error when using "gemini-3.5-flash." However, we did not encounter any errors when switching to "gemini-3.1-flash-lite." For your reference, I am attaching the gist here.
Please try using the latest SDK version.
Thanks

[github-actions]

This issue has been marked as stale because it has been open for 7 days with no activity. It will be closed in 2 days if no further activity occurs.