503 service unavailable happen randomly. Not sure why?
Is anyone else experiencing similar 503 errors with Gemini API recently?
Yes, we’re seeing the same thing — frequent 503s and occasional 502s on gemini-2.5-flash at production volume, not just occasional spikes.
Observed logs:
Error 1: 503 Service Unavailable (High Demand)
Model: gemini-2.5-flash
Endpoint: generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent
HTTP Status: 503
Latency: 1463ms
Message: [GoogleGenerativeAI Error]: Error fetching from endpoint: [503 Service Unavailable] This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later.
Error 2: 502 Bad Gateway
Model: gemini-2.5-flash
Same endpoint as above.
HTTP Status: 502
Latency: 5605ms
Message: [GoogleGenerativeAI Error]: Error fetching from endpoint: [502 Bad Gateway]
Both are on the v1beta endpoint. The 502 has a much longer timeout (5.6s) before failing than the 503 (1.46s), which suggests a routing/gateway issue on Google’s side rather than pure model overload.
We already do backoff-and-retry per the pinned guidance, but at our volume that’s not enough — retries just add latency and still fail during sustained high-demand windows. Is there an actual enterprise/capacity option for the direct Gemini Developer API, similar to Provisioned Throughput on Vertex AI? Our project is currently on Basic support so we can’t file a technical case — any guidance here would help a lot.
los errores 503 en la API de Gemini están ocurriendo últimamente debido a sobrecarga de los modelos, no por fallos de tu configuración. Se trata de un problema de capacidad temporal en la infraestructura de Google, y la solución recomendada es reintentar con exponential backoff o usar un modelo alternativo.