AI multimodality often focuses on visual, image, and video generation. Now, audio is taking center stage.

On Thursday, Google unveiled Gemini 3.1 Flash Live, which it claims to be its “highest-quality audio and voice model yet.” With the model, Google said it aims to provide a fast, natural-sounding solution for the “next generation of voice-first AI.”

Gemini 3.1 Flash Live is available to everyone in Gemini Live and Search Live, where it will help deliver more helpful and natural responses in user interactions. With the new model, Gemini Live should also be able to deliver faster responses and follow the conversation twice as long. Its multilingual capabilities are expanding Search Live into more than 200 countries and territories.

Enterprises can access the model in Gemini Enterprise for Customer Experience, and developers can via the Gemini Live API in Google AI Studio. Google claims Gemini 3.1 Flash Live is more reliable for developers and enterprises. Improvements include:

• Benchmark performance: The model outperforms the previous model on ComplexFuncBench Audio, achieving 90.8% vs 71.5%. It also topped Scale AI’s Audio MultiChallenge against leading models from OpenAI and Qwen.
• Natural Dialogue: It can better understand tone for more natural dialogues, especially useful for detecting users’ frustration or confusion.

Verizon, LiveKit and The Home Depot use the model in their workflows and have given positive feedback, Google said in its blog post.

Thursday was also a big day for open-source text-to-speech models. Cohere launched Cohere Transcribe, its first audio model. Cohere said its state-of-the-art open-source automatic speech recognition model ranks first in accuracy on HuggingFace’s Open ASR Leaderboard.

Meanwhile, Mistral launched its Voxtral TTS, an open-weight text-to-speech model that is three times smaller than the industry standard, natural-sounding, fast and supports nine languages. Naturally, the company also boasts that it has launched the best open-source text-to-speech model to date.

Our Deeper View

The fact that three new cutting-edge audio models launched on the same day is not merely a coincidence —rather, it’s representative of a broader industry trend. As AI speeds up workflows, people want to work faster. Realistically, most people can talk much faster than they can type, so they rely on transcription tools. For example, Wispr Flow, a voice-dictation tool, claims its users can be three to four times faster by dictating instead of typing. Companies are now racing to meet this demand by launching advanced models that leapfrog one another's capabilities to feed a growing user appetite.