We are currently experiencing ongoing issues with the Gemini API on a Tier 3 project (production environment), and we’re looking for clarification and guidance.
Issue Summary
For the past several days, we’ve been seeing recurring:
503 (Service Unavailable)
429 (Too Many Requests)
These errors are happening consistently and are significantly impacting production usage.
Observations
Errors started a few days ago and have been persisting without resolution
No official communication or incident report found so far
The issue appears intermittent but frequent enough to disrupt service
Error spikes are clearly visible in usage dashboards (attached screenshots)
Occurring even when traffic patterns remain relatively stable
Impact
Production degradation
Failed requests at scale
Unreliable API behavior despite being within expected usage patterns
Questions
Is there an ongoing incident or degradation affecting Gemini API (Tier 3)?
Are these 429s expected (rate limiting changes?) or unintended?
Are the 503 errors related to backend instability or capacity issues?
Is there an ETA for resolution?
Any recommended mitigation strategies on our side?
Additional Context
Tier: 3
Timeframe: last 7 days (also visible over 28 days trend)
Models used: Gemini 2.5 Flash / Flash Lite
Happy to provide more logs or request IDs if needed.
This is a critical production issue, so any visibility would be greatly appreciated.
GoogleGenerativeAIFetchError: [GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-lite:generateContent: [503 Service Unavailable] This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later.
We’re facing these errors for over a month now (approx. 30% of our requests fail and we have to retry another 3 or 4 times before it works), with no answer whatsoever until now. I’ve probably answered more than 10 posts and they never replied to me.
Besides that, if you check the Gemini Models in OpenRouter, they all show a very concerning uptime with no status update at the official API status page from google (for both Google AI Studio and for Vertex).
Thanks for the reply.
The issue is still ongoing on our side and is affecting a production Tier 3 project with recurring 503 and 429 errors over several consecutive days.
This does not appear to be isolated transient throttling, as the error spikes are significant and visible directly in the API dashboard despite relatively stable traffic patterns.
Could someone from the Gemini/API infrastructure team please confirm:
whether there is an active backend degradation,
if rate limiting behavior has recently changed,
and whether there is any ETA or mitigation guidance for production customers?
It’s clearly just computing restraints. They don’t have enough GPUs, Electricity, etc. to keep a steady uptime for all customers (and their TPUs probably have a lot of restrictions too).
The problem is the lack of transparency and acting like everything is normal, show wrong uptimes in their official API status page, etc.
We still have no clear answer or visibility regarding the ongoing Gemini API instability affecting production Tier 3 projects.
For several weeks now, we have been experiencing persistent 503 and 429 errors across Gemini 2.5 Flash / Flash Lite in a real production environment. The impact is significant and ongoing.
At this stage, what is most concerning is not only the instability itself, but the complete lack of communication and transparency around it.
We are paying customers running production workloads on your infrastructure. We need:
Clear acknowledgement of the issue
Visibility on root cause
Realistic ETA for resolution
Proper status communication when degradation occurs
If the issue is related to backend capacity or insufficient computing resources, this needs to be addressed internally and communicated professionally to developers depending on the platform.
Right now, the API behavior is unreliable despite stable traffic patterns and normal usage behavior on our side.
We are happy to provide logs, request IDs, and additional technical details if needed, but we need actual answers and visibility from the Gemini team.
The 503’s are indeed at a platform level. The best advice we can offer is:
If possible, move some traffic to the * Batch API.
Use exponential backoff and retry
For the 429’s, I’d suggest checking all the charts on your AIS rate limits page. A common issue for example is exceeding Requests Per Minute for short periods of time or exceeding search grounding limits.
(The drop in request volume over the last few days reflects the migration we already started toward another API service due to the unreliability of Gemini.)
