Audio modelPublic preview
Voxtral Mini Transcribe 3
Mistral AI
- Released
- October 20, 2026
- Data date
- October 24, 2026
Voxtral Mini Transcribe 3 transcribes audio offline and supports timestamps, speaker segmentation, and context vocabulary. The audio overview lists 65,536 shared tokens for audio and vocabulary; the model card gives a rounded 32k. The exact shared limit remains unresolved.
Specifications and access
| Specification | Value and source |
|---|---|
| API model ID | voxtral-mini-transcribe-3-0Source |
| Release | Public preview since October 20, 2026Source |
| Parameter count | 4 billionSource |
| Access | Mistral API, public previewSource |
| Context statements | Audio overview: 65,536 tokens shared by audio and vocabulary; model card: 32k, roundedSource |
| Input | Audio; vocabulary can guide recognition biasSource |
| Output | Transcript with timestamps and speaker segmentationSource |
| Release | 2026-10-20Source |
| Source status | public previewSource |
| Parameters | 4 BSource |
| Context values by source | 65,536 (audio overview); 32k, rounded (model card)Source |
| Price per audio minute in USD | 0.0035Source |
| Cached price per audio minute in USD | 0.00035Source |
| Omi report edition | Public data identifies edition 2026-10-18.1, 1,513 English recordings, 7.2096 audio hours and suite 2.0.0.Source |
| Model ID in the Omi evaluation | Internal Omi profile key voxtral-mini-transcribe-3. It is not an established API model or checkpoint identifier. (The page does not identify this as the exact Mistral API model ID and exposes no request metadata.)Source |
| Omi reference corpus | 1,513 recordings; 7.2096 hours; 63,815 reference words; 2,872 medical-term references; 442 drug-name references; 94 dosage/unit reference events.Source |
| Omi model row date | 2026-10 accuracy; batch timing 2026-10-18. Accuracy is dated only to the month; the timing run has a separate day-level date.Source |
| Language support in the current documentation | The language table lists voxtral-mini-transcribe-3-0 with support for 50 languages.Source |
| Model ID in the API examples | Official request docs show the model string voxtral-mini-transcribe-3. (The model card uses voxtral-mini-transcribe-3-0; the request documentation omits trailing '-0' and does not explain whether they are aliases.)Source |
| Artificial Analysis listed price per 1,000 minutes | 3.5Source |
| Artificial Analysis data release date | 2026-10-20Source |
| Artificial Analysis evaluation methodology | Approximately eight benchmark hours split among AgentTalk, VoxPopuli and Earnings22; datasets duration-weighted internally, then weighted 50/25/25.Source |
| Rows in the Danish results | 26,780Source |
| Danish evaluation hardware | NVIDIA A100 80 GB, batch size 16Source |
| Danish evaluation submission date | 2026-10-21Source |
Benchmark sources at a glance
Each evaluator measures different capabilities under its own conditions. Test series, composite indices and their components stay separate. The detailed results below retain their scales, configurations and source dates.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Coverage is not a quality ranking. Overlapping tests and multiple settings do not count as independent experiments. Vendor claims, historical and disputed results are excluded from this overview and remain labelled in the detailed evidence.
Page 1 of 4
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
Diarization benchmark cpWER (Show measurement, test conditions, and source)
- Source value
- 18.6
- Score
- 18.6
- Metric
- cpWER
- Unit
- %
- Category
- speaker diarization
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral Voxtral 3 announcement and benchmark tables
- Status
- active
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Macro-average over AMI-IHM, AMI-SDM, Switchboard, Switchboard CallHome, and AISHELL-4. Thirty-minute AMI and AISHELL variants are used.
Artificial Analysis weighted WER index (Show measurement, test conditions, and source)
- Source value
- 2.22
- Score
- 2.22
- Metric
- weighted WER
- Unit
- %
- Category
- speech recognition
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral Voxtral 3 announcement and benchmark tables
- Status
- active
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Weighted WER index across VoxPopuli, Earnings22, and AgentTalk.
Omi Medical batch: Medical-term WER (Show measurement, test conditions, and source)
- Source value
- 1.22
- Score
- 1.22
- Metric
- Medical-term WER
- Unit
- %
- Benchmark version
- Omi report edition 2026-10-18.1
- Category
- medical-transcription
- Direction
- Lower is better
- Sample
- 2872
- Source type
- vendor-reported
- Evaluator
- Omi Medical benchmark v1.1
- Status
- disputed
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Rank 8 in Omi's batch table. Batch transcription on the 1,513-recording English medical corpus. Rank 8; Omi internal row profile voxtral-mini-transcribe-3. Metric denominator: 2872.
- Context
- Omi labels its report edition 2026-10-18.1 and Voxtral accuracy rows only by month (2026-10); batch timing is dated October 18 and realtime timing October 19, before Mistral announced public preview on October 20. The exact endpoint/build is not identified. These are disputed observations, not definitive measurements of the post-release model.
Omi Medical batch: Overall WER (Show measurement, test conditions, and source)
- Source value
- 5.69
- Score
- 5.69
- Metric
- Overall WER
- Unit
- %
- Benchmark version
- Omi report edition 2026-10-18.1
- Category
- medical-transcription
- Direction
- Lower is better
- Sample
- 63815
- Source type
- vendor-reported
- Evaluator
- Omi Medical benchmark v1.1
- Status
- disputed
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Rank 8 in Omi's batch table. Batch transcription on the 1,513-recording English medical corpus. Rank 8; Omi internal row profile voxtral-mini-transcribe-3. Metric denominator: 63815.
- Context
- Omi labels its report edition 2026-10-18.1 and Voxtral accuracy rows only by month (2026-10); batch timing is dated October 18 and realtime timing October 19, before Mistral announced public preview on October 20. The exact endpoint/build is not identified. These are disputed observations, not definitive measurements of the post-release model.
Omi Medical batch: Drug-name errors (Show measurement, test conditions, and source)
- Source value
- 1.36
- Score
- 1.36
- Metric
- Drug-name errors
- Unit
- %
- Benchmark version
- Omi report edition 2026-10-18.1
- Category
- medical-transcription
- Direction
- Lower is better
- Sample
- 442
- Source type
- vendor-reported
- Evaluator
- Omi Medical benchmark v1.1
- Status
- disputed
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Rank 8 in Omi's batch table. Batch transcription on the 1,513-recording English medical corpus. Rank 8; Omi internal row profile voxtral-mini-transcribe-3. Metric denominator: 442.
- Context
- Omi labels its report edition 2026-10-18.1 and Voxtral accuracy rows only by month (2026-10); batch timing is dated October 18 and realtime timing October 19, before Mistral announced public preview on October 20. The exact endpoint/build is not identified. These are disputed observations, not definitive measurements of the post-release model.
Omi Medical batch: Dose/unit matches (Show measurement, test conditions, and source)
- Source value
- 79.79
- Score
- 79.79
- Metric
- Dose/unit matches
- Unit
- %
- Benchmark version
- Omi report edition 2026-10-18.1
- Category
- medical-transcription
- Direction
- Higher is better
- Sample
- 94
- Source type
- vendor-reported
- Evaluator
- Omi Medical benchmark v1.1
- Status
- disputed
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Rank 8 in Omi's batch table. Batch transcription on the 1,513-recording English medical corpus. Rank 8; Omi internal row profile voxtral-mini-transcribe-3. Metric denominator: 94.
- Context
- Omi labels its report edition 2026-10-18.1 and Voxtral accuracy rows only by month (2026-10); batch timing is dated October 18 and realtime timing October 19, before Mistral announced public preview on October 20. The exact endpoint/build is not identified. These are disputed observations, not definitive measurements of the post-release model.
Omi Medical batch: Dose/unit errors (Show measurement, test conditions, and source)
- Source value
- 20.21
- Score
- 20.21
- Metric
- Dose/unit errors
- Unit
- %
- Benchmark version
- Omi report edition 2026-10-18.1
- Category
- medical-transcription
- Direction
- Lower is better
- Sample
- 94
- Source type
- vendor-reported
- Evaluator
- Omi Medical benchmark v1.1
- Status
- disputed
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Rank 8 in Omi's batch table. Batch transcription on the 1,513-recording English medical corpus. Rank 8; Omi internal row profile voxtral-mini-transcribe-3. Metric denominator: 94.
- Context
- Omi labels its report edition 2026-10-18.1 and Voxtral accuracy rows only by month (2026-10); batch timing is dated October 18 and realtime timing October 19, before Mistral announced public preview on October 20. The exact endpoint/build is not identified. These are disputed observations, not definitive measurements of the post-release model.
Omi Medical batch: Dose precision (Show measurement, test conditions, and source)
- Source value
- 90.36
- Score
- 90.36
- Metric
- Dose precision
- Unit
- %
- Benchmark version
- Omi report edition 2026-10-18.1
- Category
- medical-transcription
- Direction
- Higher is better
- Sample
- 83
- Source type
- vendor-reported
- Evaluator
- Omi Medical benchmark v1.1
- Status
- disputed
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Rank 8 in Omi's batch table. Batch transcription on the 1,513-recording English medical corpus. Rank 8; Omi internal row profile voxtral-mini-transcribe-3. Metric denominator: 83. Precision denominator is the number of predicted dose/unit events.
- Context
- Omi labels its report edition 2026-10-18.1 and Voxtral accuracy rows only by month (2026-10); batch timing is dated October 18 and realtime timing October 19, before Mistral announced public preview on October 20. The exact endpoint/build is not identified. These are disputed observations, not definitive measurements of the post-release model.
Omi Medical batch: Dose F1 (Show measurement, test conditions, and source)
- Source value
- 84.75
- Score
- 84.75
- Metric
- Dose F1
- Unit
- %
- Benchmark version
- Omi report edition 2026-10-18.1
- Category
- medical-transcription
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Omi Medical benchmark v1.1
- Status
- disputed
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Rank 8 in Omi's batch table. Batch transcription on the 1,513-recording English medical corpus. Rank 8; Omi internal row profile voxtral-mini-transcribe-3. F1 combines reference and predicted events.
- Context
- Omi labels its report edition 2026-10-18.1 and Voxtral accuracy rows only by month (2026-10); batch timing is dated October 18 and realtime timing October 19, before Mistral announced public preview on October 20. The exact endpoint/build is not identified. These are disputed observations, not definitive measurements of the post-release model.
FLEURS 45 forced transcription WER (Show measurement, test conditions, and source)
- Source value
- 4.5
- Score
- 4.5
- Metric
- WER
- Unit
- %
- Category
- speech recognition
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral Voxtral 3 announcement and benchmark tables
- Status
- active
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Macro-average across 45 FLEURS languages, forced transcription mode. WER is used except CER for Japanese, Korean, Thai, and Chinese.
FLEURS 45 inferred transcription WER (Show measurement, test conditions, and source)
- Source value
- 4.6
- Score
- 4.6
- Metric
- WER
- Unit
- %
- Category
- speech recognition
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral Voxtral 3 announcement and benchmark tables
- Status
- active
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Macro-average across 45 FLEURS languages, inferred transcription mode. WER is used except CER for Japanese, Korean, Thai, and Chinese.
FLEURS 35 offline baseline WER (Show measurement, test conditions, and source)
- Source value
- 4
- Score
- 4
- Metric
- WER
- Unit
- %
- Category
- speech recognition / realtime comparison
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral Voxtral 3 announcement and benchmark tables
- Status
- active
- Retrieved at
- 2026-10-24
- Methodology
- Vendor-reported result. Offline Voxtral Mini Transcribe 3 baseline on FLEURS 35, as shown in the realtime transcription latency chart.
Page 1 of 4
Published prices
- Price
- USD 0.0035 per audio minute; cached rate USD 0.00035 per minuteSource
Sources and data date
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 24, 2026. 100 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Unresolved | Context length Note The audio overview lists 65,536 tokens shared by audio and vocabulary; the Mini model card gives only a rounded 32k. The exact shared context length remains unresolved. |
| Not published | Model weights and license Note The public API model card does not state downloadable weights or an open-source license. |
| Not found in the inspected sources | FLEURS results by language Note The announcement publishes macro-averages for FLEURS rather than per-language scores. |
| Unresolved | Exact Omi endpoint and model build Note Omi names the product and an internal profile key but not the exact API endpoint or model metadata. The timing conflict prevents a definitive match to the public version announced on October 20. |
| Unresolved | Model card to request ID crosswalk Note The model card uses voxtral-mini-transcribe-3-0, while the request example uses voxtral-mini-transcribe-3. Public documentation does not say whether the identifiers are interchangeable. |
| Not published | Artificial Analysis per-dataset utterance counts and WER confidence intervals Note The page/methodology reports roughly eight benchmark hours but no exact utterance counts or uncertainty interval for the WER values. |
| Additional source | Mistral Voxtral 3 announcement (retrieved October 24, 2026) · Mistral Voxtral 3 announcement · Editorial description reviewed October 24, 2026 |
| Additional source | Mistral audio models overview (retrieved October 24, 2026) · Mistral audio models overview · Editorial description reviewed October 24, 2026 |
| Additional source | Mistral speech transcription overview (retrieved October 24, 2026) |
| Additional source | Mistral model changelog (retrieved October 24, 2026) |
| Additional source | Mistral inference pricing (retrieved October 24, 2026) |
| Additional source | Voxtral Mini Transcribe 3 model card (retrieved October 24, 2026) · Voxtral Mini Transcribe 3 model card · Editorial description reviewed October 24, 2026 |
| Additional source | Omi Medical benchmark v1.1 (retrieved October 24, 2026) |
| Additional source | Artificial Analysis (retrieved October 24, 2026) |
| Additional source | Open Danish ASR Leaderboard (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |
| Additional source | Model metadata (retrieved October 24, 2026) |