Atlas

Models

← All models

Gemini 3.1 Flash Live Preview

Google · 2026-03-26

Gemini 3.1 Flash Live Preview is Google's March 2026 native-audio model for real-time, multi-turn voice conversations through the Gemini Live API. Its model card describes it as the live audio variant of Gemini 3.1 Flash Audio, based on Gemini 3 Pro; it accepts audio, images, video and text, returns audio and text, and supports configurable thinking levels such as minimal and high. The Gemini API lists a 131,072-token input limit and a 65,536-token output limit.

Benchmark scores

BenchmarkScore
Audio MultiChallenge (Audio Output)36.1
Loading Atlas data…