As the topic says, legacy API shows details errors like this when testing the API setup:
Where as the interactions API shows this
Then one has to got to the logs figure things out,
DiscourseAi::Completions::Endpoints::GeminiInteractions: status: 503 - body: {"error":{"message":"gemini-3.8-flash is currently experiencing high demand, spikes in demand are usually temporary. Please try again later.","code":"service_unavailable"}}