Thanks — you’re asking whether Gemini 3.1 can be used in LiveKit for live (real-time audio/video) sessions. I’ll answer what’s supported and what options you have next. Short answer • Gemini 3.1 models (for standard LLM/inference use) are available in LiveKit via LiveKit Inference. • Gemini 3.1 is NOT confirmed as supported for real-time “live” input/output (Gemini Live API) in the documentation. The Gemini Live examples and plugin reference a a Gemini 2.5 live model (

gemini-2.5-flash-native-audio-preview-12-2025

) for real-time audio/video. What this means for you • If your goal is to run text-based or non-realtime inference inside LiveKit (chat, generation, etc.), you can use Gemini 3.1 model IDs via LiveKit Inference. • If your goal is low-latency, real-time audio/video streaming integration (Gemini Live-style realtime sessions), Gemini 3.1 is not explicitly shown as available for that mode — the documented live examples use a Gemini 2.5 live model. Next steps I can help with • If you want to use Gemini 3.1 for non-realtime inference, I can show example configuration and model IDs to call from LiveKit Inference. • If you specifically need realtime audio/video interaction, tell me whether you can accept Gemini 2.5 for now, or whether you need Gemini 3.1. If you need 3.1 realtime, I can: ◦ Check for any updated model availability or rollout notes (and confirm if it becomes available for Gemini Live). ◦ Help you prototype with the Gemini 2.5 realtime flow so you can evaluate latency/behavior while we check 3.1 availability. Which direction would you like to take? Sources: Google Gemini LLM | LiveKit Documentation | Large language models (LLM) overview | LiveKit Documentation | LiveKit Inference | LiveKit Documentation | Gemini Live API plugin | LiveKit Documentation | Realtime models overview | LiveKit Documentation, with interactive elements