Google ships Gemini 3.1 Flash Live for lower-latency voice agents and global Search Live

On Mar 26, 2026, Google introduced Gemini 3.1 Flash Live, its latest real-time audio model built for more natural and reliable voice interaction. The company positioned it as a production-oriented model rather than a demo feature, with improvements in latency, turn-taking, tone understanding, and overall conversational stability. The goal is to make voice-first AI feel less like a sequence of audio prompts and more like an actual live exchange.

The rollout spans several product layers. Developers can access Gemini 3.1 Flash Live in preview through the Gemini Live API, while enterprises can use it through Gemini Enterprise for Customer Experience. On the consumer side, Google is also pushing the same voice capabilities into Search Live and Gemini Live. According to the announcement, Search Live is expanding to over 200 countries and territories and is getting better at adapting to a user's preferred language and local context.

Google also attached benchmark claims to the release. It said the model reaches 90.8% on ComplexFuncBench Audio and 36.1% on Audio MultiChallenge with thinking on, compared with weaker results from Gemini 2.5 Flash. The company further said the model handles tone, speaking style, noisy environments, and accent variation more reliably. Within Gemini Live, Google reported that conversations are now roughly twice as long on average, with smoother interruption handling and better correction flow when users restate or refine a request.

Trust and provenance are part of the package as well. Google said all audio generated by 3.1 Flash Live is marked with SynthID watermarking. That matters because the company is no longer presenting voice as a side capability. It is turning speech into a core interface for search, assistants, and customer experience systems, which raises both usability expectations and misuse concerns. Watermarking is one of the clearer signs that Google expects this model to be used at scale.

The broader significance is competitive. Voice is re-emerging as a primary AI surface, and the market is shifting from text-only assistants to real-time, multilingual, tool-connected agents. For developers, Gemini 3.1 Flash Live could reduce the friction of building voice products that need reliable interaction quality. For enterprises, it points to faster adoption of conversational search, support automation, and hands-free task flows across global customer bases.