I pay $250/month for the so-called “Ultra” tier, and lately I can barely use the service at all because every other request ends with:
“Our servers are experiencing high traffic right now, please try again in a minute.”
A minute? It happens repeatedly for hours.
And this is supposedly the highest consumer tier Google offers. “Ultra 20x” means nothing if paying customers are effectively rate-limited into unusability during peak times.
The most frustrating part is that things were actually better before.
Gemini 3.0 Flash Was Imperfect But Usable
Gemini 3.0 Flash had issues, yes, but at least:
-
it was responsive,
-
it was affordable in token usage,
-
and it actually worked consistently.
Then Google introduced 3.5 Flash and somehow:
-
token usage exploded,
-
costs effectively became massively higher,
-
context handling became less efficient,
-
and safety filtering became absurdly aggressive.
Now the model over-refuses constantly. Perfectly legitimate technical tasks get blocked because the safety system seems completely detached from real-world developer use cases.
It feels like:
-
ask a server question → refusal,
-
ask debugging questions → refusal,
-
ask infrastructure questions → refusal,
-
ask automation questions → “safety concerns.”
At some point it stops being an AI assistant and becomes a random refusal generator.
The Support Experience Is Worse Than the Downtime
What makes this truly infuriating is the support response pattern.
Every thread gets the same generic corporate replies:
-
“We understand your frustration.”
-
“The team is looking into it.”
-
“We appreciate your feedback.”
-
“Please submit feedback through the app.”
Meanwhile nothing changes.
No transparency.
No proper status communication.
No acknowledgement that Ultra subscribers are paying enterprise-level pricing for consumer-grade reliability.
If Google cannot provide stable capacity for Ultra users, then:
-
stop advertising these plans as premium,
-
stop pretending usage limits are meaningful,
-
and stop silently degrading access while still charging full price.
The Biggest Irony
The irony is that many developers WANT to use Gemini because:
-
the context window is excellent,
-
multimodal capabilities are promising,
-
and some outputs are genuinely impressive.
But reliability matters more than benchmarks.
An AI model that refuses half the tasks and spends the other half throwing “high traffic” errors is not production-ready.
Right now it feels like Google is prioritizing marketing announcements over actual usability.
And before someone says “it’s high demand”:
that excuse stops working when people are paying $250/month.
At that price point, users expect:
-
stable access,
-
priority compute,
-
functional support,
-
and honest communication.
