We integrate the Gemini Live API into a third-party desktop application for real-time voice conversations. We don’t control the model backend, so we can only report client-side symptoms and our usage dashboard data. We’re experiencing conversations cutting off mid-sentence/mid-response, response delays exceeding 2 minutes at times, and essentially no fully “clean” session in the past weeks.
There’s a sharp rise in errors starting Aug 27: mostly 409 Conflict (peak 40) and 429 TooManyRequests (peak 32), plus a later wave of 400 BadRequest. Notably we’re not seeing significant 503/500 errors, so this doesn’t look like a simple capacity/overload issue on the surface, though happy to be corrected if that’s not the right read.
Models affected: Gemini 3 Flash Live, Gemini 3.1 Flash Lite