Tasks and inputs
Voxtral Small Transcribe 3 is listed for Automatic speech recognition, Speech → Text. 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 Small Transcribe 3 is Mistral's top-accuracy batch speech-to-text model for recorded audio, with context biasing and 50+ languages.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative: $0.004 per audio minute. 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
Recorded calls, depositions, earnings webcasts. When the transcript is the deliverable and nobody is waiting on it live, accuracy matters more than turnaround, and Voxtral Small Transcribe 3 is the tier Mistral built for that job. It sits at the top of the Voxtral 3 speech family, went live on 20 October 2026, and is billed by the audio minute.
It is a closed model. Earlier Voxtral checkpoints shipped open weights; this one is listed in Mistral's Premier tier as version 3.0, with no weights to download.
| Item | Detail |
|---|---|
| API ID | voxtral-small-transcribe-3-0 |
| Released | 20 October 2026 |
| Input / output | Audio in, text out |
| Batch languages | 50+, including all 24 official EU languages |
| Indicative price | About $0.004 per audio minute |
| Weights | Closed (Premier) |
| Hosting | EU-hosted API, with US and global endpoints |
Domain words are where general transcription falls apart. Drug names, ticker symbols, a client's product codes. Instead of fine-tuning, you send context biasing with the request: a list of key terms, or a paragraph of free-form text describing what the recording is about. One deployment can then serve many customers, each with its own glossary, and nobody waits on a training run.
The language claim is broad. Beyond the EU set, the Transcribe 3 family is aimed at a range of lower-resource Asian and Indic languages too. If that is your market, test it with your own recordings rather than taking the headline count on trust.
Here the published material disagrees with itself. The launch messaging presents this model as doing transcription and diarization, meaning who spoke when. The audio documentation says it supports transcription only at launch. If speaker labels are essential, send a two-speaker test file and check whether the response actually carries them before designing around it. When they are missing, the smaller Voxtral Mini Transcribe 3 is documented with speaker labels and segment timestamps, and it costs slightly less.
Anyone still on Voxtral Mini Transcribe 2, deprecated on 31 October 2026, will be landing on one of these three.
Pull fifty real files, not clean benchmark audio. Include crosstalk, accents, phone-line compression and the jargon your users actually say. Run every file through Small and Mini with the same biasing list. Then measure error rates on the terms that carry business meaning, separately from overall word error rate, because a model can win the average and still mangle the one product name that matters.
Then do the arithmetic. At these rates the premium over Mini is half a tenth of a cent per minute: trivial for legal review, noticeable at a million minutes a month.
Prices are indicative; confirm them in your own Mistral console before committing. The rest of the lineup is on the Mistral provider page, and alternatives from other labs sit under automatic speech recognition.
Compare the model on the work you need to complete.
Voxtral Small Transcribe 3 is listed for Automatic speech recognition, Speech → Text. It accepts audio and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
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 3 is Mistral's 4B batch transcription model with speaker labels, segment timestamps and custom vocabulary, replacing Mini 2.
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.