What is Voxtral 3 (Transcribe and TTS)?
Voxtral 3 is Mistral AI's speech family, released in public preview on 20 October 2026. It has four models: Voxtral Small Transcribe 3 and the 4-billion-parameter Voxtral Mini Transcribe 3 for recorded audio, with speaker labels (diarization) and timestamps; Voxtral Mini Transcribe Realtime 3 for live transcription in under 500 milliseconds; and Voxtral TTS 3, which speaks text with zero-shot voice cloning and streaming. Mistral launched a managed voice-agent pipeline and SDK the same day.
Transcription covers 50 languages, including all 24 official EU languages, and speech output covers more than 35 languages. The models are available through Mistral's API with EU, US and global endpoints. Unlike some earlier Voxtral releases, no weights have been published; Mistral offers self-deployment only as an enterprise arrangement.
Reported launch prices are $0.0035 a minute for Mini Transcribe 3, $0.0040 for Small Transcribe 3, $0.008 for the real-time model and $13 per million characters for speech.
How much does Voxtral 3 (Transcribe and TTS) cost?
| Plan | Price | What you get |
|---|---|---|
| Mini Transcribe 3 | $0.0035 a minute | Recorded audio, diarization and timestamps |
| Small Transcribe 3 | $0.0040 a minute | |
| Mini Transcribe Realtime 3 | $0.008 a minute | Live transcription under 500 ms |
| TTS 3 | $13 per million characters | Zero-shot voice cloning, streaming |
Mini Transcribe 3 $0.0035 a minute; Small Transcribe 3 $0.0040 a minute; Mini Transcribe Realtime 3 $0.008 a minute; TTS 3 $13 per million characters (launch prices as reported).
Is Voxtral 3 (Transcribe and TTS) worth it?
Voxtral 3's transcription is the reason to look. Artificial Analysis measured 2.1% word error rate for Small Transcribe 3, 0.1 points behind Microsoft's MAI-Transcribe-2 and 0.1 ahead of ElevenLabs Scribe v2, and 2.2% for the cheaper Mini. At $0.0035 a minute, an hour of audio costs about 21 cents, against $0.22 an hour for Scribe v2, so it matches the leaders on accuracy and price.
Mistral's own numbers back up the speaker-labelling: 18.4% diarization error for Small, ahead of Pyannote's 19.0% and far ahead of Gemini 3.5 Transcribe's 38.4%. The real-time model trades accuracy for speed predictably, from 5.7% error at 2.4 seconds of delay to 7.4% at 240 milliseconds, so voice agents should test where their latency budget lands.
The speech side is less of a standout. Voxtral TTS 3 sits 13th on Artificial Analysis's speech leaderboard at $13 per million characters, roughly 70 cents per hour of speech. It is 127 Elo points better than Mistral's open 4B TTS from March, but behind ElevenLabs' Eleven v4.
Two things to weigh. All four models are in public preview, so prices and behaviour may change. And unlike Voxtral Realtime and Voxtral 4B TTS earlier this year, there are no downloadable weights: teams that need offline speech should stay with those older open releases.
Where it earns its place
A European call-centre analytics company
Transcribe and label speakers on thousands of hours of calls
Small Transcribe 3 handles diarization and timestamps through Mistral's EU endpoint. 10,000 hours cost about $2,400 at $0.0040 a minute.
A team building a phone voice agent
Hear callers in real time and answer with a natural voice
Pair Mini Transcribe Realtime 3 with TTS 3 in Mistral's voice-agent SDK. Set the transcription delay around 480 to 960 ms to keep error near 6%.
A podcast or video publisher
Produce accurate multilingual transcripts and subtitles
Mini Transcribe 3 at $0.0035 a minute matches Small within 0.1 points on Artificial Analysis; a 60-minute episode costs about 21 cents.
Before you commit
- Setup
- Call voxtral-mini-transcribe-3-0, voxtral-small-transcribe-3-0, voxtral-mini-transcribe-realtime-3-0 or voxtral-mini-tts-3-0 on Mistral's API, choosing the EU, US or global endpoint.
- Integrations
- Mistral API and Voice Agents pipeline; LiveKit-based voice-agent SDK covering listening, reasoning, tool calls and speech.
- Data and privacy
- API only; no published weights or licence. Mistral's API terms let it train on free-tier data unless you opt out, with opt-out for pay-as-you-go; check them for audio before sending sensitive calls. Self-deployment is an enterprise arrangement.
- Cost at scale
- 100,000 hours of transcription cost about $21,000 on Mini or $48,000 on real-time. Speech output at $13 per million characters is about 70 cents per hour of audio.
- Limits
- Public preview. Real-time error rises from 5.7% at 2.4 s delay to 7.4% at 240 ms. Speech output in 35+ languages; transcription in 50.
What's good
- 2.1% word error rate on Artificial Analysis, among the top three
- About 21 cents an hour of audio on Mini Transcribe 3
- Diarization error 18.4%, ahead of Pyannote and Gemini, per Mistral
- EU, US and global endpoints; 50 transcription languages
What's not
- No open weights, unlike earlier Voxtral releases
- Public preview; prices and models may change
- TTS 3 ranks 13th on Artificial Analysis's speech board
- Real-time accuracy drops to 7.4% error at 240 ms delay
Best for
Teams transcribing large volumes of meetings, calls or media, especially in Europe, who want top-tier accuracy at low per-minute cost.
Voxtral 3 (Transcribe and TTS) alternatives
Voxtral 3 (Transcribe and TTS): common questions
- Is Voxtral 3 (Transcribe and TTS) worth it?
- Near the top of independent transcription tests at a third of a cent a minute, with EU endpoints; the speech side is merely competitive and nothing is open-weight this time. We rate it 4.4 out of 5.
- How much does Voxtral 3 (Transcribe and TTS) cost?
- Mini Transcribe 3 $0.0035 a minute; Small Transcribe 3 $0.0040 a minute; Mini Transcribe Realtime 3 $0.008 a minute; TTS 3 $13 per million characters (launch prices as reported).
- What are the downsides of Voxtral 3 (Transcribe and TTS)?
- No open weights, unlike earlier Voxtral releases; public preview; prices and models may change; tTS 3 ranks 13th on Artificial Analysis's speech board; real-time accuracy drops to 7.4% error at 240 ms delay.
- Who is Voxtral 3 (Transcribe and TTS) best for?
- Teams transcribing large volumes of meetings, calls or media, especially in Europe, who want top-tier accuracy at low per-minute cost.
- How much does Voxtral 3 (Transcribe and TTS) cost for a team or in production?
- 100,000 hours of transcription cost about $21,000 on Mini or $48,000 on real-time. Speech output at $13 per million characters is about 70 cents per hour of audio.
- Is Voxtral 3 (Transcribe and TTS) safe for company data?
- API only; no published weights or licence. Mistral's API terms let it train on free-tier data unless you opt out, with opt-out for pay-as-you-go; check them for audio before sending sensitive calls. Self-deployment is an enterprise arrangement.
- What are the best Voxtral 3 (Transcribe and TTS) alternatives?
- The closest alternatives we have reviewed are ElevenLabs Scribe v2, Gemini 3.5 Transcribe, Deepgram Nova-3 and GPT Transcribe.
Answers checked Oct 25, 2026.
Make Voxtral 3 (Transcribe and TTS)? Show this rating on your site
