I am building a workflow that requires an agent to run parallel calls to Gemini (I am using gemini-3-flash-preview) right now. I have Google AI Ultra subscription for precisely this reason - to be able to get dedicated resources to develop my agent. Not only am I facing issues with Antigravity IDE erroring out due to “servers being too busy”, even the calls to Gemini API by my agent are getting timed out. I haven’t had a single successful agent run in 24 hours - all requests timing out since some nodes succeed, some nodes nodes do not. And this is when I downgraded all nodes to use Gemini 3 Flash. In reality I need some of them to use Gemini 2.5/3/3.1 Pro. But again - too frequent timeouts. My only recourse is to go back to Gemini 2.5 Flash or 1.5 Flash to at least continue my dev work, and hope that they don’t time out this frequently.
Suggestion - you should really consider tiered resourcing. Ultra subscribers pay a lot of money for specific needs, which unfortunately are not being met with a broken experience. My experience has been degrading daily - first only Antigravity IDE started throwing errors (once or twice) which has now increased to every 9/10 retries. And now Gemini API timing out. This experience does not justify the price we pay. Ultra subscribers should get dedicated resources and more availability of resources. I am seriously considering migrating my AI stack to Anthropic - models with better coding skills, and less frequent timeouts for now.