Google said on September 15 that it was launching Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two voice AI models available that day through the Gemini API, Google Workspace, and the Gemini app.

The models support spoken interactions, can use visual input, and can carry out background tasks during a conversation.

Google said the models can handle interruptions and switch between 97 supported languages mid-conversation. The company added that Gemini 3.8 Live Extended Thinking could narrate its progress while working through multi-step requests.

Google also said audio generated by its AI products was watermarked with SynthID, and that it had published a model card with safety details.

The watermarking move is consistent with a European Union policy push to help users identify AI-generated content, and with AI Act transparency obligations meant to distinguish synthetic media.

🔗 Source: Google

Recent Google developments