AI News Roundup: ElevenLabs Realtime, Voxtral 3, and OpenAI on Google Cloud

Here is the AI news from the last day that matters for teams building agents and voice workflows. Every item links to its source.

ElevenLabs launches Realtime v1 Mini and v1 Max

ElevenLabs released two speech-to-speech models that handle listening, reasoning, and speaking in one system. Both support more than 90 languages. Mini is the lower-cost option for assistants, while Max is aimed at agents that must follow a process, such as customer support and qualifying prospects. The company also says its interface works with code written for OpenAI's Realtime API. (The Economic Times)

Mistral releases Voxtral 3 and a voice agent SDK

Mistral AI shipped Voxtral 3, a family of transcription and speech models paired with a LiveKit-based kit for phone and web agents. Agents can look things up and take actions during a call, like booking an appointment, and Mistral says turns complete in under one second. (TAO Media)

Google Cloud adds OpenAI models to Model Garden

Google Cloud began rolling out GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna in private preview, so customers can pick and swap models on one invoice. (Google Cloud Blog)

Gradium open-sources Phonon for on-device speech

Gradium released Phonon, a 31-million-parameter text-to-speech model that runs locally on phones, laptops, and in the browser, under the Apache 2.0 license. (Gradium)

Voice agents keep getting faster and better at following a script, which is exactly what a sales call needs. If you want an AI voice agent that calls your leads, qualifies them, and hands ready buyers to a human closer, see Ozkra or get in touch.