What is Voxtral Transcribe 2?
Voxtral Transcribe 2 is two models that share 13 languages. Voxtral Mini Transcribe 2 works on recorded files through Mistral's API: it labels who spoke, times each word, takes up to 100 hint words for names and jargon, and accepts up to 3 hours of audio per request. Voxtral Realtime writes text while audio streams in, with a delay you choose, and its 4B-parameter weights are free under Apache 2.0.
Mistral has deprecated both. They keep working until they retire at the end of April 2027, and new integrations are pointed to Voxtral 3, whose Small model covers 50 languages and whose live model adds speaker labels, though none of it has open weights.
What replaces Voxtral Transcribe 2?
Voxtral 3: Mistral has deprecated both Transcribe 2 models and points each one to a Voxtral 3 model.
- Recorded files
- Voxtral Mini Transcribe 3 costs $0.0035 a minute; Voxtral Small Transcribe 3 costs $0.004 and covers 50 languages.
- Live audio
- Voxtral Mini Transcribe Realtime 3 costs $0.008 a minute and adds speaker labels, end-of-turn detection and a delay from 80 ms to 2.4 seconds.
- Open weights
- Voxtral 3 has none so far; the Transcribe 2 Realtime weights stay available under Apache 2.0.
- Timeline
- Both Transcribe 2 models keep working until they retire at the end of April 2027; after that, requests fail.
Who is Voxtral Transcribe 2 best for?
Voxtral Transcribe 2 is best for developers who already run its low-cost, speaker-labelled transcripts, or who want an open live model to host themselves.
Meeting, interview and call transcripts
Speaker labels with start and end times show who said what, and one request can take a 3-hour recording.
Self-hosted live transcription
Voxtral Realtime's Apache 2.0 weights run on a single GPU with 16GB or more and stay usable after the API retires, which Voxtral 3 cannot offer.
Low-cost transcripts in its languages
At $0.003 a minute, or $0.18 an hour, batch transcripts cost less than Voxtral Mini Transcribe 3 at $0.0035 a minute and ElevenLabs Scribe v2 at $0.22 an hour.
Teams already running it
Existing integrations keep working until the end of April 2027, which leaves time to test Voxtral 3 on your own audio before switching.
What's included with Voxtral Transcribe 2?
Speaker labels, context biasing, word timestamps, an adjustable live delay and open weights for the live model.
- Speaker labels
- Batch transcripts mark each speaker with start and end times, though when people talk over each other it usually keeps only one voice.
- Context biasing
- Up to 100 words or phrases steer the spelling of names and jargon; it is tuned for English and experimental in the other languages.
- Word and segment timestamps
- Timings for every word or segment suit subtitles and audio search, but they cannot be combined with a fixed language setting.
- Adjustable live delay
- Voxtral Realtime streams over a WebSocket; Mistral says a 480 ms delay keeps accuracy close to the batch model and a 2.4-second delay matches it.
- Open weights for the live model
- The 4B Realtime weights are on Hugging Face under Apache 2.0 and run in vLLM or Transformers.
How much does Voxtral Transcribe 2 cost?
Voxtral Transcribe 2 costs $0.003 per minute of audio for batch transcription and $0.006 per minute for live transcription on Mistral's API until both models retire, and Mistral's Free plan no longer includes API credits.
- Batch transcription in 13 languages
- Speaker labels, word timestamps and context biasing
- Up to 3 hours of audio per request
- Live transcription with a delay you set
- 13 languages, 4B parameters
- Open weights under Apache 2.0 on Hugging Face
Mistral has deprecated both models and no longer lists them on its API pricing page; these prices come from their model cards, and both stay available until they retire at the end of April 2027. Mistral's Free plan no longer includes API credits, so API use is paid from prepaid credits or from the monthly API credits in paid plans such as Pro ($14.99 a month). Voxtral Realtime can also run free on your own hardware. The replacements cost $0.0035 a minute (Voxtral Mini Transcribe 3) and $0.008 a minute (Voxtral Mini Transcribe Realtime 3). Prices are in US dollars. Full pricing on docs.mistral.ai. We checked these prices on .
How do readers rate Voxtral Transcribe 2?
No one has rated Voxtral Transcribe 2 yet. Be the first.
Sign in to rate. One rating per account.
Voxtral Transcribe 2 pros and cons
Users praise its accuracy and low price, but Mistral has deprecated it, it covers only 13 languages, and it keeps speaker labels to the closed batch model.
Pros
- Hacker News and Reddit testers found it accurate on fast speech, jargon, noisy audio and heavy UK accents; one team rated it level with Deepgram Nova-3 on its own call recordings.
- Batch transcripts cost $0.18 an hour of audio, below ElevenLabs Scribe v2's $0.22, and users describe the API as fast, with one 20-minute file back in 18 seconds.
- The live model's Apache 2.0 weights allow commercial self-hosting, and developers ported it to C, Rust and Apple's MLX within a week of release.
Cons
- Deprecated: Mistral points new integrations to Voxtral 3, has dropped both models from its API pricing page, and plans to retire them at the end of April 2027.
- Only 13 languages, and speech outside them snaps to the nearest supported one: users got Russian text for Ukrainian and Polish, and Hindi for Bengali.
- Speaker labels work only on the batch model, which has no open weights; the live model cannot label speakers or detect the end of a turn, both of which Voxtral Mini Transcribe Realtime 3 now does.
- The open model is heavy for local use: Mistral's card lists a 16GB GPU for serving it, and several testers found it too slow on laptops and CPUs next to lighter models such as NVIDIA Parakeet.
What are the best Voxtral Transcribe 2 alternatives?
Voxtral 3 for new integrations, Scribe v2 for 90+ languages, and Gemini 3.5 Transcribe for a free tier.
Voxtral 3PaidChoose Voxtral 3 for any new integration: it is Mistral's replacement, with 50 languages on Voxtral Small Transcribe 3 at $0.004 a minute and a live model with speaker labels and end-of-turn detection at $0.008, though without open weights.
Scribe v2FreeChoose it when you need languages beyond Voxtral's 13; ElevenLabs covers 90+ languages for $0.22 an hour of audio, a little more than Voxtral's $0.18.
Gemini 3.5 TranscribeFreeChoose it when you want a free tier and 85+ languages; paid use costs about $0.005 a minute, but files stop at 1 hour, or 30 minutes with speaker labels.
Questions about Voxtral Transcribe 2
What is Voxtral Transcribe 2?
Voxtral Transcribe 2 is Mistral AI's pair of speech-to-text models for developers: one writes speaker-labelled transcripts of recordings, the other writes live text with open weights. Mistral now points users to Voxtral 3.
Who is Voxtral Transcribe 2 for?
Voxtral Transcribe 2 is best for developers who already run its low-cost, speaker-labelled transcripts, or who want an open live model to host themselves.
Is Voxtral Transcribe 2 free?
Only the live model's weights. Voxtral Realtime runs free on your own hardware under Apache 2.0, but Mistral's Free plan no longer includes API credits, so API use is paid: $0.003 a minute for batch and $0.006 a minute for live transcription.
How much does Voxtral Transcribe 2 cost?
Voxtral Transcribe 2's paid plans are Voxtral Mini Transcribe 2 at $0.003 per minute of audio and Voxtral Mini Transcribe Realtime at $0.006 per minute of audio.
What are the best Voxtral Transcribe 2 alternatives?
The closest alternatives on Recaply are Voxtral 3, Scribe v2 and Gemini 3.5 Transcribe.
Where can I try Voxtral Transcribe 2?
On its official website, mistral.ai.
Is Voxtral Transcribe 2 being discontinued?
Yes. Mistral has deprecated Voxtral Mini Transcribe 2 and Voxtral Realtime and plans to retire both at the end of April 2027, after which API calls to them fail. Mistral points new integrations to Voxtral 3, while the Realtime weights stay on Hugging Face under Apache 2.0.
Which languages does Voxtral Transcribe 2 support?
Voxtral Transcribe 2 handles 13 languages: English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian and Dutch. Context biasing, which steers the spelling of names and terms, is tuned for English and still experimental in the others.
Can I run Voxtral Transcribe 2 locally?
Only the live model. Voxtral Realtime's 4B weights are on Hugging Face under Apache 2.0 and run in vLLM or Transformers on a single GPU with 16GB or more. Voxtral Mini Transcribe 2, the batch model with speaker labels, is available only through Mistral's API.
What model name do I use for Voxtral Transcribe 2 in the API?
For recorded files, send audio to the /v1/audio/transcriptions endpoint with voxtral-mini-2602; the alias voxtral-mini-latest still points to it. For live audio, open a WebSocket session with voxtral-mini-transcribe-realtime-2602. Names containing transcribe-3 or transcribe-latest call Voxtral 3 instead.
Our verdict
Voxtral Transcribe 2 is still one of the cheapest ways to get accurate, speaker-labelled transcripts in its 13 languages, but Mistral has deprecated it.
Batch transcripts cost $0.003 a minute and testers found them accurate on jargon, noise and strong accents, yet both models retire at the end of April 2027 and Voxtral 3 adds languages and live speaker labels.
- Skip it for
- New projects, which should start on Voxtral 3, and anyone who works in languages outside the 13.
- Check first
- If you already use it, run your own recordings through Voxtral Mini Transcribe 3 before the retirement date.
Cohere Transcribe
Voxtral TTS
MAI-Transcribe-2-Streaming
Eleven v4
AssemblyAI-TTS-1
Gemini 3.8 Flash TTS