I’m running an n8n workflow that uses the Google Gemini node (gemini-2.5-flash-lite model) with the Google Search option enabled, all through the Google AI Studio free tier.
Here’s what the workflow does:
The Gemini node is prompted to search for today’s 10 trending topics across the web, plus 5 RSS feeds and 5 specific websites. It then returns a compiled list of the top 10 trending topics.
The issue:
Every time I run the Gemini node, I get a 503 error saying the model is currently experiencing high demand. I’ve tried running it at different times, but the error persists. However, when I reduced the request to just 3 topics and removed the RSS feeds and specific websites, the same model (gemini-2.5-flash-lite) ran without any errors.
My questions are:
1. If I set up billing and start paying, would I still run into the same issue?
2. Why does the same request work perfectly fine on the Gemini website but always fail when called via the API?
503 = model overloaded on Google’s side, not really your request size, but bigger prompts (10 topics + 10 sources + Search grounding) take longer, so you’re more likely to hit a transient overload window than a tiny 3-topic request.
Yes, paying helps, free tier shares capacity with everyone and gets deprioritized under load; a billing-enabled project gets much better availability, especially with gemini-2.5-flash-lite which is popular and often congested.
Website vs API differs because AI Studio’s web UI has its own retry/queueing behind the scenes, the raw API just returns 503 immediately and expects you to retry.
Fix regardless of tier: add retry-with-backoff on the Gemini node (n8n has built-in retry settings, enable “Retry On Fail” with a few seconds delay), and consider splitting the 10-source Search-grounded call into smaller chunks if 503s persist even on billing.