MiniMaxAudio modelActive
MiniMax Speech 2.8
PopularityRank 670 of 875
0out of 100same as last month
Summary
MiniMax Speech 2.8 controls expressive speech synthesis through native sound tags and voice cloning from a ten-second reference. The family produces multilingual output and explicitly names Mandarin and Japanese among its supported languages.
The API offers the identifiers speech-2.8-hd and speech-2.8-turbo. MiniMax lists $100 per 1M characters for HD and $60 for Turbo for both synchronous and asynchronous speech synthesis. Voice design and cloning are billed separately.
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.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, Chinese (zh), CER (Show measurement, test conditions, and source)
- Source value
- 3.42
- Score
- 3.42
- Metric
- character error rate (CER) on CV3-Eval multilingual voice cloning, Chinese (zh)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, Chinese (zh), CER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, English (en), WER (Show measurement, test conditions, and source)
- Source value
- 3.45
- Score
- 3.45
- Metric
- word error rate (WER) on CV3-Eval multilingual voice cloning, English (en)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, English (en), WER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, Japanese (ja), CER (Show measurement, test conditions, and source)
- Source value
- 6.29
- Score
- 6.29
- Metric
- character error rate (CER) on CV3-Eval multilingual voice cloning, Japanese (ja)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, Japanese (ja), CER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, Korean (ko), CER (Show measurement, test conditions, and source)
- Source value
- 7.49
- Score
- 7.49
- Metric
- character error rate (CER) on CV3-Eval multilingual voice cloning, Korean (ko)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, Korean (ko), CER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, German (de), WER (Show measurement, test conditions, and source)
- Source value
- 3.3
- Score
- 3.3
- Metric
- word error rate (WER) on CV3-Eval multilingual voice cloning, German (de)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, German (de), WER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, Spanish (es), WER (Show measurement, test conditions, and source)
- Source value
- 2.79
- Score
- 2.79
- Metric
- word error rate (WER) on CV3-Eval multilingual voice cloning, Spanish (es)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, Spanish (es), WER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, French (fr), WER (Show measurement, test conditions, and source)
- Source value
- 8.74
- Score
- 8.74
- Metric
- word error rate (WER) on CV3-Eval multilingual voice cloning, French (fr)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, French (fr), WER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, Italian (it), WER (Show measurement, test conditions, and source)
- Source value
- 3.67
- Score
- 3.67
- Metric
- word error rate (WER) on CV3-Eval multilingual voice cloning, Italian (it)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, Italian (it), WER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, Russian (ru), WER (Show measurement, test conditions, and source)
- Source value
- 5.39
- Score
- 5.39
- Metric
- word error rate (WER) on CV3-Eval multilingual voice cloning, Russian (ru)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, Russian (ru), WER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, Arabic (ar), WER (Show measurement, test conditions, and source)
- Source value
- 3.38
- Score
- 3.38
- Metric
- word error rate (WER) on CV3-Eval multilingual voice cloning, Arabic (ar)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, Arabic (ar), WER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, Indonesian (id), WER (Show measurement, test conditions, and source)
- Source value
- 1.46
- Score
- 1.46
- Metric
- word error rate (WER) on CV3-Eval multilingual voice cloning, Indonesian (id)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, Indonesian (id), WER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
Qwen-Audio-3.0-TTS paper Table 4: MiniMax-Speech-2.8-HD, CV3-Eval multilingual voice-cloning subset, Portuguese (pt), WER (Show measurement, test conditions, and source)
- Source value
- 1.86
- Score
- 1.86
- Metric
- word error rate (WER) on CV3-Eval multilingual voice cloning, Portuguese (pt)
- Unit
- %
- Benchmark version
- arXiv:2607.23938v1, published 2026-07-27; evaluation date not stated
- Category
- audio
- Direction
- Lower is better
- Source type
- vendor-reported
- Evaluator
- Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Exact MiniMax-Speech-2.8-HD public-API result (API model speech-2.8-hd); CV3-Eval multilingual voice-cloning subset, Portuguese (pt), WER. The paper gives no per-row evaluated sample count or confidence interval.
- Context
- This preprint is authored by Alibaba Token Foundry, a competing speech-generation provider. Treat its comparative scores as vendor-reported, not an independent third-party evaluation. They apply only to the exact model named in this row; MiniMax-Speech-2.8-HD results do not establish Speech 2.8 Turbo performance.
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
pyannote Community-1, pyannote: 73
Nemotron-3-Diarization, NVIDIA: 57
Gemini 3.5 Transcribe, Google: 52
Confucius4-R2T2, NetEase Youdao: 52
Granite Speech 5.0 470M TurboCTC, IBM: 52
Kokoro-82M, hexgrad: 51
Nova-3, Deepgram: 51
Fish Audio S2 Pro, Fish Audio: 51
NVIDIA Nemotron 3.5 Transcribe, NVIDIA: 50
Granite Speech 5.0 470M TurboCTC NC, IBM: 50
Orukeet, Oruk: 49
Lyria 3.5, Google: 47
Gemini 3.5 Live Translate, Google: 46
Phonon-2, Fermion Research: 44
Aura-2, Deepgram: 44
Gemini 3.1 Flash TTS, Google: 44
MiniMax-Music3, MiniMax: 41
Audio8-ASR-Infinite, Edge0 AI: 41
GPT-Realtime 2.1, OpenAI: 41
MiniMax Speech 2.8, MiniMax: 0
MiniMax Speech 2.8
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 5, 2026
MiniMax Speech 2.8Amazon Nova 2 SonicEleven Multilingual STS v2Eleven Music v2.5Eleven Text to Sound v2
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
- Evaluated API variant
- Exact API variant: speech-2.8-hd (retrieved 2026-10-16)
- Source: Evaluated API variant
Prices
- HD price
- speech-2.8-hd costs $100 per 1M characters, synchronous and asynchronousSource
- Turbo price
- speech-2.8-turbo costs $60 per 1M characters, synchronous and asynchronousSource
- Voice services
- Voice Design $3 and Voice Cloning $1.50 per voice, charged on first synthesis use; preview synthesis is billed separatelySource
Sources and data date
Show all 36 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.
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 4, 2026. 31 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Unresolved | quality-benchmarks-turbo Note The Qwen-Audio paper still contains no Turbo result; its reported results are HD-specific. Vapi separately publishes Humanness 93 explicitly for MiniMax Speech 2.8 Turbo. Scale, direction, evaluation N, confidence interval, and score protocol remain unknown; the 1,063 votes and 20 prompts do not establish evaluation N. The Vapi value is not a family-wide quality result, and other Turbo benchmark coverage remains unresolved. |
| Unresolved | quality-benchmarks-turbo Note Artificial Analysis results remain explicitly HD-specific. Vapi separately publishes Humanness 93 explicitly for MiniMax Speech 2.8 Turbo. Scale, direction, evaluation N, confidence interval, and score protocol remain unknown; the 1,063 votes and 20 prompts do not establish evaluation N. The Vapi value is not a family-wide benchmark, and other Turbo benchmark coverage remains unresolved. |
| Unresolved | Vapi Humanness Index scale, direction, and evaluation N Original wording The opened exact-model page publishes 93 but does not specify the score scale, direction, confidence interval, or independent evaluation sample count. It reports blind votes and a fixed prompt set, neither of which is used as sampleCount. |
| Unresolved | Vapi Speech 2.8 latency measurement details Original wording The 325 ms latency footnote names the speech-2.5-turbo-preview request ID despite the page's Speech 2.8 Turbo claim; exact tested model for the latency result remains unresolved. |
| Additional source | MiniMax Speech pricing (retrieved October 3, 2026; October 4, 2026) · MiniMax official pricing · Editorial description reviewed October 3, 2026 |
| Additional source | MiniMax Speech 2.8 announcement (retrieved October 3, 2026; October 4, 2026) · MiniMax Speech 2.8 announcement · Editorial description reviewed October 3, 2026 |
| Additional source | Qwen-Audio-3.0-TTS paper, Alibaba Token Foundry (retrieved October 16, 2026) |
| Additional source | Artificial Analysis (retrieved October 17, 2026; October 28, 2026; October 29, 2026; November 3, 2026) |
| Additional source | VoiceData Research (retrieved October 17, 2026) |
| Additional source | Design Arena Audio Realism leaderboard (retrieved October 17, 2026) |
| Additional source | Artificial Analysis Controlled Voice Arena (retrieved October 18, 2026; October 28, 2026; October 29, 2026; October 30, 2026; November 3, 2026) |
| Additional source | Vapi Humanness Index (retrieved October 19, 2026; October 23, 2026) |
| Additional source | DesignArena (retrieved October 28, 2026) |
| Additional source | Speko (retrieved October 28, 2026) |
| Additional source | VoiceArena primary public leaderboard (retrieved October 28, 2026) |
| Additional source | Model metadata (retrieved October 4, 2026)12 |
| Additional source | Model metadata (retrieved October 15, 2026; October 16, 2026)12 |
| Additional source | Model metadata (retrieved October 17, 2026)123 |
| Additional source | Model metadata (retrieved October 28, 2026)123456789101112 |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |