Gemini 3.8 Live models now available on AI Gateway

Vercel’s AI Gateway now hosts Google’s Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking models, expanding real-time conversational capabilities for developers. The models enable applications to process and respond to spoken input instantly, supporting voice assistants and interactive audio experiences. Both versions maintain continuous engagement by allowing background tool calls without disrupting the conversation flow.

Gemini 3.8 Live supports real-time audio input and output, visual grounding, and automatic language switching across 97 languages. It also handles background tool calls while the user speaks, ensuring uninterrupted interaction. Developers can integrate the model by minting a short-lived token and opening a WebSocket connection for real-time event handling.

Gemini 3.8 Live Extended Thinking introduces parallel multi-step reasoning that runs alongside speech output. This feature lets the model acknowledge requests and narrate progress without pausing the conversation, enhancing responsiveness in dynamic interactions. The model adapter serializes and parses real-time events for seamless integration.

AI Gateway provides a unified API for model access, usage tracking, and cost management. It includes features for configuring retries, failover, and performance optimizations to improve uptime beyond provider guarantees. Developers can test the models in the AI Gateway model playground before deployment.