---
title: 'Mistral ships Voxtral 3 and ElevenLabs ships Eleven Realtime as voice agents become a full stack race'
url: 'https://securteksecurity.com/news/mistral-voxtral-3-elevenlabs-eleven-realtime-voice-agents-race'
markdown: 'https://securteksecurity.com/news/mistral-voxtral-3-elevenlabs-eleven-realtime-voice-agents-race.md'
date: '2026-10-20'
description: 'Mistral launched Voxtral 3 on October 20, 2026, a family built for real-time voice agents: a speech-to-text model, Voxtral Small Transcribe 3, covering more than 50 languages, a speech model, Voxtral Mini TTS 3, supporting more than 35 languages and more than 50 dialects with emotion, pacing and em…'
taxonomy:
  category:
    - AI
---

# [Mistral ships Voxtral 3 and ElevenLabs ships Eleven Realtime as voice agents become a full stack race](https://securteksecurity.com/news/mistral-voxtral-3-elevenlabs-eleven-realtime-voice-agents-race)

   20th Oct 2026  

Mistral launched Voxtral 3 on October 20, 2026, a family built for real-time voice agents: a speech-to-text model, Voxtral Small Transcribe 3, covering more than 50 languages, a speech model, Voxtral Mini TTS 3, supporting more than 35 languages and more than 50 dialects with emotion, pacing and emphasis as instructions, and a voice agent SDK based on LiveKit that streams listening, reasoning, tool calls and speech in one pipeline. Agents can tap external systems during live conversations, for example to book appointments or check invoices. The system can run through EU-hosted APIs or be self-deployed, and transcription pricing starts at $0.0035 per minute.

ElevenLabs shipped two speech-to-speech models the same day, Eleven Realtime v1 Max and v1 Mini, through ElevenAgents and an OpenAI-compatible Realtime API. ElevenLabs says v1 Max scores 98.8 percent on the Big Bench Audio benchmark with a 0.57 second response time, company figures that have not yet been independently benchmarked; v1 Mini is a cheaper model built for natural conversation and can tell when a speaker has finished or is only saying "mm-hmm." The same day timing from both companies moves voice competition from transcription accuracy to turn latency, interruptibility and ownership of the whole agent loop.

Sources: [The Decoder, Mistral](https://the-decoder.com/mistral-ai-launches-voxtral-3-a-full-voice-agent-stack-that-listens-thinks-and-talks-in-real-time/); [The Decoder, ElevenLabs](https://the-decoder.com/elevenlabs-releases-two-new-speech-to-speech-models-designed-for-real-time-ai-conversations/); [CryptoBriefing](https://cryptobriefing.com/mistral-unveils-voxtral-3-voice-agent-sdk/); [Oossa](https://oossa.com/en/elevenlabs-launches-two-real-time-speech-to-speech-models)

---

## Navigation

- Parent: [News](https://securteksecurity.com/news.md)
- Previous: [Mirae Asset Life Insurance breach leaks customer personal and credit information in South Korea](https://securteksecurity.com/news/mirae-asset-life-insurance-breach-south-korea.md)
- Next: [Moldova government breach extracts data linked to about 750,000 people](https://securteksecurity.com/news/moldova-government-breach-data-linked-to-750000-people.md)