Mistral AIAudio modelActive
Voxtral Mini TTS
PopularityRank 474 of 876
29out of 100same as last month
Summary
Voxtral Mini TTS combines text-to-speech with zero-shot voice cloning. A voice sample is enough for the voice prompt, and Mistral says no transcript of that sample is required. The model supports nine languages and produces first audio in about 90 ms during streaming according to the provider. This combination defines the endpoint for voices that need to start quickly from a clone.
Mistral lists the Voxtral Mini TTS weights under CC BY-NC 4.0. The license applies whether the checkpoint runs locally or through the API. API access and local weights are separate access paths, but neither bypasses the non-commercial terms.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 86 published benchmark results
Page 1 of 8
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.
Voxtral TTS paper human preference win rate versus ElevenLabs Flash v2.5 (zero-shot voice cloning) (Show measurement, test conditions, and source)
- Source value
- 68.4
- Score
- 68.4
- Metric
- Micro-average human preference win rate across nine languages
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 5
- Category
- audio
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact open-weight checkpoint mistralai/Voxtral-4B-TTS-2603. The paper's Table 5 reports the micro-average human preference win rate against ElevenLabs Flash v2.5 in zero-shot voice cloning across Arabic, Dutch, English, French, German, Hindi, Italian, Portuguese, and Spanish. The study compares generated audio with reference clips on likeness, naturalness, and expressivity. The paper does not report the aggregate vote denominator or a confidence interval.
- Context
- This is a Mistral-authored, vendor-reported human preference result for zero-shot voice cloning, not an independent MOS or a general TTS preference score. The paper gives study setup details but no total scored pairwise comparisons, vote denominator, or confidence interval, so this record omits a sample count and confidence bounds.
Voxtral TTS Table 3 MiniMax-TTS Arabic WER (Show measurement, test conditions, and source)
- Source value
- 2.68
- Score
- 2.68
- Metric
- Word error rate (WER) of generated speech, measured by Voxtral Mini Transcribe v2
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 3, MiniMax-TTS Arabic
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact released checkpoint mistralai/Voxtral-4B-TTS-2603; Voxtral column in Table 3 for MiniMax-TTS Arabic. The paper measures WER with Voxtral Mini Transcribe v2. Table 3's WER and UTMOS values match Table 6's DPO values for the same task; the paper says DPO changes speaker similarity by at most 0.01.
- Context
- Mistral-authored automatic result; WER is transcribed by Voxtral Mini Transcribe v2, so it is not an independent transcription audit. The paper does not report the scored sample count, run date, or confidence interval for this row.
Voxtral TTS Table 3 MiniMax-TTS German WER (Show measurement, test conditions, and source)
- Source value
- 0.83
- Score
- 0.83
- Metric
- Word error rate (WER) of generated speech, measured by Voxtral Mini Transcribe v2
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 3, MiniMax-TTS German
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact released checkpoint mistralai/Voxtral-4B-TTS-2603; Voxtral column in Table 3 for MiniMax-TTS German. The paper measures WER with Voxtral Mini Transcribe v2. Table 3's WER and UTMOS values match Table 6's DPO values for the same task; the paper says DPO changes speaker similarity by at most 0.01.
- Context
- Mistral-authored automatic result; WER is transcribed by Voxtral Mini Transcribe v2, so it is not an independent transcription audit. The paper does not report the scored sample count, run date, or confidence interval for this row.
Voxtral TTS Table 3 MiniMax-TTS English WER (Show measurement, test conditions, and source)
- Source value
- 0.63
- Score
- 0.63
- Metric
- Word error rate (WER) of generated speech, measured by Voxtral Mini Transcribe v2
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 3, MiniMax-TTS English
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact released checkpoint mistralai/Voxtral-4B-TTS-2603; Voxtral column in Table 3 for MiniMax-TTS English. The paper measures WER with Voxtral Mini Transcribe v2. Table 3's WER and UTMOS values match Table 6's DPO values for the same task; the paper says DPO changes speaker similarity by at most 0.01.
- Context
- Mistral-authored automatic result; WER is transcribed by Voxtral Mini Transcribe v2, so it is not an independent transcription audit. The paper does not report the scored sample count, run date, or confidence interval for this row.
Voxtral TTS Table 3 MiniMax-TTS Spanish WER (Show measurement, test conditions, and source)
- Source value
- 0.51
- Score
- 0.51
- Metric
- Word error rate (WER) of generated speech, measured by Voxtral Mini Transcribe v2
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 3, MiniMax-TTS Spanish
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact released checkpoint mistralai/Voxtral-4B-TTS-2603; Voxtral column in Table 3 for MiniMax-TTS Spanish. The paper measures WER with Voxtral Mini Transcribe v2. Table 3's WER and UTMOS values match Table 6's DPO values for the same task; the paper says DPO changes speaker similarity by at most 0.01.
- Context
- Mistral-authored automatic result; WER is transcribed by Voxtral Mini Transcribe v2, so it is not an independent transcription audit. The paper does not report the scored sample count, run date, or confidence interval for this row.
Voxtral TTS Table 3 MiniMax-TTS French WER (Show measurement, test conditions, and source)
- Source value
- 3.22
- Score
- 3.22
- Metric
- Word error rate (WER) of generated speech, measured by Voxtral Mini Transcribe v2
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 3, MiniMax-TTS French
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact released checkpoint mistralai/Voxtral-4B-TTS-2603; Voxtral column in Table 3 for MiniMax-TTS French. The paper measures WER with Voxtral Mini Transcribe v2. Table 3's WER and UTMOS values match Table 6's DPO values for the same task; the paper says DPO changes speaker similarity by at most 0.01.
- Context
- Mistral-authored automatic result; WER is transcribed by Voxtral Mini Transcribe v2, so it is not an independent transcription audit. The paper does not report the scored sample count, run date, or confidence interval for this row.
Voxtral TTS Table 3 MiniMax-TTS Hindi WER (Show measurement, test conditions, and source)
- Source value
- 4.99
- Score
- 4.99
- Metric
- Word error rate (WER) of generated speech, measured by Voxtral Mini Transcribe v2
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 3, MiniMax-TTS Hindi
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact released checkpoint mistralai/Voxtral-4B-TTS-2603; Voxtral column in Table 3 for MiniMax-TTS Hindi. The paper measures WER with Voxtral Mini Transcribe v2. Table 3's WER and UTMOS values match Table 6's DPO values for the same task; the paper says DPO changes speaker similarity by at most 0.01.
- Context
- Mistral-authored automatic result; WER is transcribed by Voxtral Mini Transcribe v2, so it is not an independent transcription audit. The paper does not report the scored sample count, run date, or confidence interval for this row.
Voxtral TTS Table 3 MiniMax-TTS Italian WER (Show measurement, test conditions, and source)
- Source value
- 1.32
- Score
- 1.32
- Metric
- Word error rate (WER) of generated speech, measured by Voxtral Mini Transcribe v2
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 3, MiniMax-TTS Italian
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact released checkpoint mistralai/Voxtral-4B-TTS-2603; Voxtral column in Table 3 for MiniMax-TTS Italian. The paper measures WER with Voxtral Mini Transcribe v2. Table 3's WER and UTMOS values match Table 6's DPO values for the same task; the paper says DPO changes speaker similarity by at most 0.01.
- Context
- Mistral-authored automatic result; WER is transcribed by Voxtral Mini Transcribe v2, so it is not an independent transcription audit. The paper does not report the scored sample count, run date, or confidence interval for this row.
Voxtral TTS Table 3 MiniMax-TTS Dutch WER (Show measurement, test conditions, and source)
- Source value
- 1.99
- Score
- 1.99
- Metric
- Word error rate (WER) of generated speech, measured by Voxtral Mini Transcribe v2
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 3, MiniMax-TTS Dutch
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact released checkpoint mistralai/Voxtral-4B-TTS-2603; Voxtral column in Table 3 for MiniMax-TTS Dutch. The paper measures WER with Voxtral Mini Transcribe v2. Table 3's WER and UTMOS values match Table 6's DPO values for the same task; the paper says DPO changes speaker similarity by at most 0.01.
- Context
- Mistral-authored automatic result; WER is transcribed by Voxtral Mini Transcribe v2, so it is not an independent transcription audit. The paper does not report the scored sample count, run date, or confidence interval for this row.
Voxtral TTS Table 3 MiniMax-TTS Portuguese WER (Show measurement, test conditions, and source)
- Source value
- 1.02
- Score
- 1.02
- Metric
- Word error rate (WER) of generated speech, measured by Voxtral Mini Transcribe v2
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 3, MiniMax-TTS Portuguese
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact released checkpoint mistralai/Voxtral-4B-TTS-2603; Voxtral column in Table 3 for MiniMax-TTS Portuguese. The paper measures WER with Voxtral Mini Transcribe v2. Table 3's WER and UTMOS values match Table 6's DPO values for the same task; the paper says DPO changes speaker similarity by at most 0.01.
- Context
- Mistral-authored automatic result; WER is transcribed by Voxtral Mini Transcribe v2, so it is not an independent transcription audit. The paper does not report the scored sample count, run date, or confidence interval for this row.
Voxtral TTS Table 3 SEED-TTS WER (Show measurement, test conditions, and source)
- Source value
- 1.23
- Score
- 1.23
- Metric
- Word error rate (WER) of generated speech, measured by Voxtral Mini Transcribe v2
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 3, SEED-TTS
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact released checkpoint mistralai/Voxtral-4B-TTS-2603; Voxtral column in Table 3 for SEED-TTS. The paper measures WER with Voxtral Mini Transcribe v2. Table 3's WER and UTMOS values match Table 6's DPO values for the same task; the paper says DPO changes speaker similarity by at most 0.01.
- Context
- Mistral-authored automatic result; WER is transcribed by Voxtral Mini Transcribe v2, so it is not an independent transcription audit. The paper does not report the scored sample count, run date, or confidence interval for this row.
Voxtral TTS Table 4 explicit steering win rate vs ElevenLabs v3 (Show measurement, test conditions, and source)
- Source value
- 51
- Score
- 51
- Metric
- Pairwise human preference win rate, ties excluded, explicit emotion steering
- Unit
- %
- Benchmark version
- Voxtral TTS paper arXiv:2603.25551v2, Table 4, flagship voices, explicit steering
- Category
- audio
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Mistral AI, Voxtral TTS paper
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact released checkpoint mistralai/Voxtral-4B-TTS-2603; Table 4 compares the four flagship voices (British female, British male, American male, French female) with competitor voices matched for gender and accent. Voxtral uses another reference clip from the same speaker that conveys the requested emotion; it accepts no emotion tags or text instructions. Gemini 2.5 Flash TTS receives free-form instructions, and ElevenLabs v3 receives emotion tags. Three native speakers of the same dialect annotate each language/model pair; the reported win rate excludes ties.
- Context
- Mistral-authored pairwise human preference result. The study describes 77 prompts and three native-speaker annotators of the same dialect per language/model pair, but not the total scored comparisons, vote denominator, or confidence interval. Table 4 excludes ties. This measures preference under the stated steering setup, not standardized emotion recognition.
Page 1 of 8
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
0 to 100, higher is more popular
20 of 83 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 6, 2026
Voxtral Mini TTSIndex-Echo-S2ST-2B 30MOSS-Audio-4B-Thinking 29Index-Echo-S2ST-9B 29Scribe v2 Realtime 29
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Voxtral Mini TTS compared
- 1Voxtral Mini TTS vs. Eleven v33 shared tests
- 2Voxtral Mini TTS vs. Gemini 3.1 Flash TTS2 shared tests
- 3Voxtral Mini TTS vs. GPT-4o Mini TTS1 shared test
- 4Voxtral Mini TTS vs. Sonic 3.62 shared tests
- 5Voxtral Mini TTS vs. Fish Audio S2 Pro2 shared tests
- 6Voxtral Mini TTS vs. Eleven Flash v2.52 shared tests
- 7Voxtral Mini TTS vs. Eleven v43 shared tests
- 8Voxtral Mini TTS vs. Eleven v4 Turbo3 shared tests
- 9Voxtral Mini TTS vs. Fish Audio S2.1 Pro2 shared tests
Show all 15 comparisons
- 10Voxtral Mini TTS vs. Gemini 3.8 Flash-Lite TTS2 shared tests
- 11Voxtral Mini TTS vs. Gemini 3.8 Flash TTS3 shared tests
- 12Voxtral Mini TTS vs. MAI-Voice-21 shared test
- 13Voxtral Mini TTS vs. MAI-Voice-2-Flash1 shared test
- 14Voxtral Mini TTS vs. MiniMax Speech 2.82 shared tests
- 15Voxtral Mini TTS vs. Voxtral Mini TTS 32 shared tests
Similar models
More models of the same kind as Voxtral Mini TTS, chosen by model line, developer, size, price and release date.
- Voxtral Mini Transcribe RealtimeMistral AI · same developer
- Voxtral Mini Transcribe 2Mistral AI · same developer
- Voxtral Mini Transcribe 3Mistral AI · same developer
- Voxtral Mini Transcribe Realtime 3Mistral AI · same developer
- Voxtral Mini TTS 3Mistral AI · same developer
- Voxtral Mini TTS Realtime 3Mistral AI · same developer
- Voxtral Small Transcribe 3Mistral AI · same developer
- pyannote Community-1pyannote · popular in this category
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
API
Further details
Prices
- Speech generation price
- $16 per 1M charactersSource
Sources and data date
Show all 22 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.