I’m a paid Tier 2 user of Gemini Enterprise Agent Platform
I’ve been running into persistent 429 Resource Exhausted errors while using the standard service tier (not flex) for Gemini 3.8 Flash, global location, single project, single agent, no parallel tasks.
It happens fairly regularly nowadays, and when it does happen it can last for hours. During these episodes, using the API becomes a game of chance–send a request and maybe I’ll get a few agentic responses before the 429 returns, or I might get one response, or even none. If I’m so lucky, I might actually get work done, but not without spending up to a few hours baby-ing my agent. Maybe my time and attention isn’t particularly valuable, but I know money is!
Did you know the difference between 80% and 90% cache rate (% of total input tokens cached) equals a 33% reduction in overall input costs?
Did you know 95% of the cost of agentic tasks can be from input tokens?
Well, my cache rate drops from 95% to 60% during these periods, tripling my total costs. Whenever I hit the 429 roadblock, my context cache gets cleared. Do it enough times, and my total cost skyrockets. But there’s no way to know whether the spike is over without trying, so I have to keep trying (with spacing, of course).
As someone who sadly depends on Gemini for timely work, all of this is extremely frustrating. When I remember I’m paying for this (tier 2!), it goes from frustrating to infuriating.
I also have a Google AI Pro subscription and have spent more than 10 hours getting bounced between Google One, Google Cloud, and Google Developers Program while trying to figure out why I’m not getting the monthly Cloud credits promised by the plan. That has been resolved, but now Jules access isn’t upgraded as promised. I’m sure there’s more if I keep trying to use the stuff I’ve been promised.
My experience with the Google ecosystem has been dismal, to say the least.
I reside in Hong Kong, where OpenAI and Anthropic don’t offer services, but I prefer US providers, which is why I picked Google. The fact that someone like me with no other choice is considering finding ways around the geofence (which is costly and involves setting up overseas companies), or even patronizing Chinese providers (while noting the risk to my code), speaks volumes about how poor my experience with Google has been.
I just wanted to get this off my chest. I hope this helps Google stay in the AI game, though I’m definitely looking around now.