Tasks and inputs
Voxtral Mini Transcribe 3 is listed for Automatic speech recognition, Speech → Text, Speaker Diarization. It accepts audio and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 17, 2026
Voxtral Mini Transcribe 3 is Mistral's 4B batch transcription model with speaker labels, segment timestamps and custom vocabulary, replacing Mini 2.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative: $0.0035 per audio minute (batch). Confirm on your own Mistral account.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
If your pipeline still calls voxtral-mini-2602, it has been running on borrowed time. Mistral deprecated Voxtral Mini Transcribe 2 on 31 October 2026, and the replacement is Voxtral Mini Transcribe 3: a 4B-parameter batch model released on 20 October that keeps the low-latency, low-price brief of its predecessor while adding the structure most transcription jobs end up needing anyway.
Three things come back with the text. Speaker labels. Segment timestamps. And respect for a custom vocabulary list you supply.
| Item | Detail |
|---|---|
| API ID (model table) | voxtral-mini-transcribe-3-0 |
| Parameters | 4B |
| Context | 65,536 tokens, shared by audio and vocabulary |
| Output | Text with speaker labels and segment timestamps |
| Indicative price | About $0.0035 per audio minute (batch) |
| Weights | Closed (Premier) |
| Replaces | voxtral-mini-2602, deprecated 31 October 2026 |
The model table lists voxtral-mini-transcribe-3-0. The Python sample in the audio docs calls voxtral-mini-transcribe-3, without the suffix. Both may resolve; we could not confirm that. Try the full ID first, and if your account rejects it, fall back to the short form and note which one worked in your config.
The call itself goes through client.audio.transcriptions.complete(...) in Mistral's Python SDK.
That 65,536-token window is shared. Audio tokens and vocabulary tokens draw from the same pool, so a sprawling glossary of five thousand product SKUs leaves less room for the recording itself. Mistral does not convert the window into minutes of audio. Find your own ceiling empirically: send progressively longer files with your real vocabulary list attached, and note where results start to truncate or degrade. Then split long recordings below that point.
Keep the vocabulary lean. Twenty terms that are genuinely misheard beat two thousand that the model already gets right.
The licence model, for one. Earlier Voxtral releases such as Voxtral Mini 3B 2507 came with downloadable weights; this one does not. If self-hosting was the reason you picked Voxtral, that option is gone at this tier, though Mistral does offer on-premises deployment of the stack through its sales team.
Need higher accuracy on hard audio? The larger Voxtral Small Transcribe 3 costs about $0.004 a minute. Need live captions? The Realtime 3 variant streams. More options are on the Mistral provider page and the speaker diarization index.
Treat the price as indicative and confirm it in your Mistral console.
Compare the model on the work you need to complete.
Voxtral Mini Transcribe 3 is listed for Automatic speech recognition, Speech → Text, Speaker Diarization. It accepts audio and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Documented limits: 65.54K context. Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
salesforce
SOV-Transcribe-1 is Salesforce's open-weight speech-augmented model for non-streaming transcription, with keyword biasing and style instructions; Apache-2.0.
ElevenLabs
Scribe v3 Live is ElevenLabs' streaming transcription model with turn prediction, mid-sentence language switching and a 10,000-term keyterm library.
NVIDIA
Nemotron 3.5 Transcribe 30B A4B is NVIDIA's open offline ASR model: 40 languages, hour-long files, word timestamps, diarisation and translation.
IBM
Granite Speech 5.0 2B is IBM's open speech-language model for multilingual ASR and translation with keyword biasing, runnable from Ollama to MLX.
Mistral AI
Voxtral Mini Transcribe Realtime 3 is Mistral's streaming speech-to-text model for voice agents: under 500 ms, tunable delay, built-in turn detection.
Mistral AI
Voxtral Small Transcribe 3 is Mistral's top-accuracy batch speech-to-text model for recorded audio, with context biasing and 50+ languages.