On October 20, 2026, Mistral AI released Voxtral 3 — a family of transcription and text-to-speech models plus a LiveKit-based voice-agent SDK that runs speech in, reasoning, and speech out as one streaming pipeline across 50+ languages.
2026-10-20 · by Moe Ameen
On October 20, 2026, Mistral AI announced Voxtral 3, a new family of speech models paired with a voice-agent SDK that runs speech-to-text, text-to-speech, and tool calling in one real-time streaming pipeline. The release includes batch transcription models — Voxtral Small Transcribe 3 and the 4B-parameter Voxtral Mini Transcribe 3 — a streaming model built for agents (Voxtral Mini Transcribe Realtime 3), and a text-to-speech model, Voxtral Speech 3. Transcription covers 50+ languages, including the 24 official EU languages; the real-time model handles 35+, and the TTS side covers 35+ languages and 50+ dialects.
Mistral priced the transcription models at $0.0035 per minute for Voxtral Mini Transcribe 3, $0.0040 per minute for Voxtral Small Transcribe 3, and $0.008 per minute for the real-time model, with text-to-speech at $13 per million characters. The full voice-agent API, which pairs the speech models with a reasoning model over the LiveKit-based SDK, is listed at $0.04 per minute with Mistral Small 4 as the reasoner or $0.08 per minute with another model. Mistral says real-time transcription runs at sub-500ms latency and that a complete agent turn finishes in under one second.
Beyond plain transcription, the models add speaker diarization (labeling who spoke when), context biasing (feeding in key terms to improve recall without fine-tuning), and end-of-turn detection that uses both acoustic and semantic cues. The TTS side adjusts emotion, pacing, and emphasis from context and supports voice cloning from sampled audio. On benchmarks, Mistral reports Voxtral Small Transcribe 3 outperforming Whisper, ElevenLabs Scribe v2, and Gemini 3.5 Transcribe and reaching state-of-the-art results on FLEURS and diarization — these are company-reported figures and are not independently verified here.
Voxtral 3 is available through Mistral's EU-hosted API plus US and global endpoints, with self-deployment possible for teams with data-sovereignty requirements (a full on-premises stack requires contacting Mistral). It builds on earlier Voxtral releases — the open 2025 speech-understanding models and the March 2026 text-to-speech model. Because most specs come from Mistral's own launch, verify current pricing and availability against Mistral before relying on any figure.
The practical shift here is not the SDK — it is that spoken content is now the cheapest raw material you own. A faster, more accurate, sub-cent transcription layer means the hour you already spent recording a podcast, a webinar, a client call, or a quick voice memo is a mine you can work, not a file that rots in a folder. [Kompozy](/) is built to do exactly that mining: it already runs Whisper-based ASR inside its render pipeline, so you never choose or wire a speech model — you hand it the audio or video and it does the transcription step for you.
From one recording, Kompozy generates captioned short clips, [Carousel](/glossary/hyperframes) posts and Quote Graphics built from the strongest lines, a blog article and an email newsletter, and captioned [Persona Shorts](/glossary/persona-shorts) in your brand voice — all held to one tone by the [Persona Brief](/glossary/persona-brief), then scheduled and published across the eight social platforms plus blog and email from one queue with [Autopilot](/glossary/autopilot). Voxtral 3 makes the words cheap to capture; Kompozy turns a single recording into a week of posts.
Voxtral 3 is Mistral AI's family of speech models — batch and real-time speech-to-text plus a text-to-speech model — paired with a LiveKit-based voice-agent SDK that runs transcription, reasoning, and speech generation as one real-time streaming pipeline. Mistral announced it on October 20, 2026.
By Mistral's launch pricing, transcription runs $0.0035/min (Voxtral Mini Transcribe 3), $0.0040/min (Voxtral Small Transcribe 3), and $0.008/min for the real-time model; text-to-speech is $13 per million characters; and the full voice-agent API is $0.04/min with Mistral Small 4 or $0.08/min with another reasoning model. Confirm current rates on Mistral's site.
Not by itself. Voxtral 3 produces an accurate transcript (and can generate speech), but it does not cut clips, write captions, design carousels, or publish anywhere. To turn that transcript into captioned shorts, carousels, a blog, a newsletter, and scheduled posts across platforms, you use a content engine like Kompozy, which handles the transcription step for you with Whisper-based ASR.