Google announced on 2026-09-15 that Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking are rolling out to developers through the Gemini API and Google AI Studio.
The launch targets real-time spoken agents. Per the Google announcement, the models can work across spoken input, text, and visual context, then respond in speech or text. Google also says Live can process visual context during a conversation and handle tool or API work in the background without pausing the exchange, a notable shift for agents that need to fetch data or stage actions mid-call.
Access still looks uneven. Google used rollout language rather than promising uniform production availability, and some enterprise paths remain private preview or coming soon. Pricing is also unsettled: independent reporting disagrees on whether published token rates apply to a distinct 3.8 Live SKU or an older Live Preview listing, so teams should confirm access and costs in their own account before scaling. Read the pricing caveat.