What happened
Mistral AI launched Voxtral 3, bundling a speech-to-text model (Voxtral Small Transcribe 3), a text-to-speech model (Voxtral Mini TTS 3) and a voice agent SDK based on LiveKit into one streaming pipeline, with transcription covering more than 50 languages.
Why it matters
The bundle lets companies build voice agents that listen, reason, call external tools and speak back in real time, and can run on their own infrastructure or through EU-hosted APIs to keep data under their control, according to Mistral.
What to watch
Text-to-speech output supports over 35 languages and 50 dialects with emotional adaptation, and pricing starts at $0.0035 per minute for transcription.
WHO IT HITSCompanies building voice agents — customer support teams, appointment-booking services and invoice-checking workflows — can now assemble a real-time voice pipeline from one vendor, potentially simplifying procurement and data-governance reviews.
Summaries like this, in your inbox every morning.
Voxtral 3 arrives as a bundle rather than a single model. Mistral has packaged a speech-to-text model (Voxtral Small Transcribe 3), a text-to-speech model (Voxtral Mini TTS 3) and an agent SDK built on LiveKit, so developers do not have to stitch together separate vendors for each stage of a voice agent. The company says the combined pipeline handles listening, reasoning, tool use and speaking in a single stream, and that agents can reach external systems mid-conversation — for example, to book appointments or check invoices.
The deployment story is equally part of the pitch. Mistral says the system can run on a customer's own infrastructure or through EU-hosted APIs, a setup aimed at keeping data under the customer's control. That option may matter to organizations in regulated sectors weighing where voice data is processed.
On specifications, Mistral says transcription covers more than 50 languages and text-to-speech output supports over 35 languages and 50 dialects with emotional adaptation. Pricing starts at $0.0035 per minute for transcription, with further details in the documentation.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
DeepSeek released DeepInfer, the inference engine behind V4.1-Flash, with production figures of 155 trillion t…
OpenAI acquired Orchestra Bio on October 20 and is integrating its team and drug-development coordination tech…
Memoket began selling the Memoket Gem on the 21st for 31,980 yen — an 11g AI recorder that mounts on an Apple…
A preliminary experiment ran 100 copies of Qwen/Qwen3-1.7B, each with its own LoRA (small add-on weights actin…
Claroc searched passive DNS for cf.metaaiusercontent.com and found more than 6,800 Muse share links, because t…
Google's "Personal Intelligence with Gemini" began rolling out on Android Auto on the 20th, and will extend to…
