gemini-live-2.5-flash-native-audio

gemini-live-2.5-flash-native-audio is an AI model from Google Gemini designed to hold real-time voice conversations with people.

Most voice AI works by first transcribing sound into text, understanding that text to form a reply, then converting the reply back into speech. It's a bit like listening to a foreign language by mentally reading subtitles, then translating your answer back before speaking. This model skips that translation step entirely — it listens to sound as sound and responds directly in sound. That means it can pick up on things text can't capture, like the tremor in someone's voice or the exact timing of an interruption, and react to them.

In the name, 'Live' refers to real-time conversation, and 'Flash' indicates a lighter, faster version of the model. Unlike text-based chatbots, it's built for use cases where actual voices are exchanged back and forth, such as voice assistants or support agents.

In articles, it shows up as a component of voice agent pipelines, in phrases like "each stage runs on the gemini-live-2.5-flash-native-audio model." It's easy to miss that this means the model processes sound input and output directly from the start, rather than simply adding a voice to a text-based reply after the fact.