Skip to main content

MiniMaxAudio modelOpen weights

MiniMax-Music3

PopularityTop group about equally popular, 8 models

50out of 100new calculation

based on two data groups, mostly search, range 32 to 68

Summary

MiniMax-Music3 generates complete songs from lyrics and a detailed music description. Its published card mentions vocals, accompaniment, editable musical structures, and 32 kHz stereo WAV output of up to five minutes.

The weights and source code are public. Since August 20, 2026, the paid music API has been closed to new users, while existing paying users can continue using it. The free music APIs are being discontinued. The local model release is separate from the API offer and MiniMax Token Plan.

GraduallyBench Audio

MiniMax-Music3

Technical test
  • Only a technical test is available.
  • The selection of test inputs is limited.
  • The route does not confirm the exact model version.
  • The functional test does not assess content quality.
Show test details

Technical measurement

fal · Language not documented

4 valid outputs from 4 attempted and 4 planned test slots.

No further scored metric is available for this test.

WER compares words and CER compares characters with the reference. The speech output was re-transcribed with GPT Transcribe. Its exact backend revision is unconfirmed. Read speech and digital silence do not cover spontaneous speech or real background noise. Error rates can exceed 100 %. Where used, the test references were not independently reviewed by humans. Test coverage and methods differ by model. Each metric stands on its own; there is no combined overall score across tests. Data compiled on 11/10/2026. How we test · Download all Decision and Audio results as JSON

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 12 published benchmark results

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.

PER (Show measurement, test conditions, and source)
Source value
6.27
Metric
phoneme error rate
Unit
%
Benchmark version
WildSongBench comparison snapshot 2026-09-12
Category
audio
Direction
Lower is better
Sample
192
Source type
vendor-reported
Evaluator
YuE2 model card
Status
active
Retrieved at
2026-10-04
Methodology
WildSongBench snapshot dated September 12, 2026: 192 selected outputs from 192 prompts. MiniMax Music 3 used its official caption rewriter. The documented system comparison is not matched for compute.
Context
The YuE2 model team published this comparison; it is not an independent evaluation.
Source YuE2 model card
A Thousand Listeners overall leaderboard preference score (Show measurement, test conditions, and source)
Source value
39.4
Metric
scenario-equalized pairwise preference score
Unit
%
Benchmark version
A Thousand Listeners report schema v1.0
Category
audio-generation
Direction
Higher is better
Lower confidence bound
36
Upper confidence bound
43.1
Source type
independent-evaluator
Evaluator
A Thousand Listeners
Status
active
Retrieved at
2026-10-22
Methodology
Independent evaluation. Blind head-to-head listening on full-song generations from 104 prompts across 13 scenarios; each scenario is weighted equally. The frozen primary cohort contains 1,000 listeners and 6,000 primary judgements. The source reports a 95% interval; uncertainty uses 300 judgement-level bootstrap resamples (seed 20260829), resampled within each scenario and then equal-weighted across all scenarios. No single independent per-model sample count is defined, so sampleCount remains unset.
Context
The report labels the evaluated MiniMax system as Music 3.0, but does not fully document checkpoint provenance beyond that label or the full generation and playback protocol. Prompts and generated lyrics are Mandarin only; the result is bounded to 13 scenarios and the recruited listener population, not general-population preference.
Source A Thousand Listeners
Musicality (Show measurement, test conditions, and source)
Source value
5.3482
Metric
score
Unit
No unit provided
Benchmark version
WildSongBench comparison snapshot 2026-09-12
Category
audio
Direction
Higher is better
Sample
192
Source type
vendor-reported
Evaluator
YuE2 model card
Status
active
Retrieved at
2026-10-04
Methodology
WildSongBench snapshot dated September 12, 2026: 192 selected outputs from 192 prompts. MiniMax Music 3 used its official caption rewriter. The documented system comparison is not matched for compute.
Context
The YuE2 model team published this comparison; it is not an independent evaluation.
Source YuE2 model card
SongBench average (Show measurement, test conditions, and source)
Source value
6.283
Metric
score
Unit
No unit provided
Benchmark version
WildSongBench comparison snapshot 2026-09-12
Category
audio
Direction
Higher is better
Sample
192
Source type
vendor-reported
Evaluator
YuE2 model card
Status
active
Retrieved at
2026-10-04
Methodology
WildSongBench snapshot dated September 12, 2026: 192 selected outputs from 192 prompts. MiniMax Music 3 used its official caption rewriter. The documented system comparison is not matched for compute.
Context
The YuE2 model team published this comparison; it is not an independent evaluation.
Source YuE2 model card
SongEval average, MiniMax Music 3 (Show measurement, test conditions, and source)
Source value
4.0993521875
Metric
SongEval mean across five dimensions
Unit
points
Benchmark version
WildSongBench September 12, 2026 comparison; results at dataset commit 4b0e5226cdb4ba08e1774d4cfd8d5fc87bb90c37
Category
audio
Direction
Higher is better
Sample
192
Source type
independent-evaluator
Evaluator
WildSongBench evaluation authors
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. Exact WildSongBench system row MiniMax Music 3. September 12, 2026 comparison; metrics use the same 192 evaluated outputs. The recorded setup uses MiniMax's official caption rewriter. This is not a matched-compute comparison.
Context
The evaluator reports these as system-level automatic metrics, not human preference. No confidence intervals are published, and systems were not evaluated with matched compute. The three pre-existing SongBench Musicality, SongBench average, and PER values were deduplicated rather than re-added.
Source WildSongBench evaluation authors
SongEval Musicality, MiniMax Music 3 (Show measurement, test conditions, and source)
Source value
4.0430729167
Metric
SongEval Musicality score
Unit
points
Benchmark version
WildSongBench September 12, 2026 comparison; results at dataset commit 4b0e5226cdb4ba08e1774d4cfd8d5fc87bb90c37
Category
audio
Direction
Higher is better
Sample
192
Source type
independent-evaluator
Evaluator
WildSongBench evaluation authors
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. Exact WildSongBench system row MiniMax Music 3. September 12, 2026 comparison; metrics use the same 192 evaluated outputs. The recorded setup uses MiniMax's official caption rewriter. This is not a matched-compute comparison.
Context
The evaluator reports these as system-level automatic metrics, not human preference. No confidence intervals are published, and systems were not evaluated with matched compute. The three pre-existing SongBench Musicality, SongBench average, and PER values were deduplicated rather than re-added.
Source WildSongBench evaluation authors
AudioBox PQ, MiniMax Music 3 (Show measurement, test conditions, and source)
Source value
8.2824510212
Metric
AudioBox production-quality score
Unit
rating
Benchmark version
WildSongBench September 12, 2026 comparison; results at dataset commit 4b0e5226cdb4ba08e1774d4cfd8d5fc87bb90c37
Category
audio
Direction
Higher is better
Sample
192
Source type
independent-evaluator
Evaluator
WildSongBench evaluation authors
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. Exact WildSongBench system row MiniMax Music 3. September 12, 2026 comparison; metrics use the same 192 evaluated outputs. The recorded setup uses MiniMax's official caption rewriter. This is not a matched-compute comparison.
Context
The evaluator reports these as system-level automatic metrics, not human preference. No confidence intervals are published, and systems were not evaluated with matched compute. The three pre-existing SongBench Musicality, SongBench average, and PER values were deduplicated rather than re-added.
Source WildSongBench evaluation authors
MuLan, MiniMax Music 3 (Show measurement, test conditions, and source)
Source value
0.3927985193
Metric
MuLan text/audio alignment score
Unit
ratio
Benchmark version
WildSongBench September 12, 2026 comparison; results at dataset commit 4b0e5226cdb4ba08e1774d4cfd8d5fc87bb90c37
Category
audio
Direction
Higher is better
Sample
192
Source type
independent-evaluator
Evaluator
WildSongBench evaluation authors
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. Exact WildSongBench system row MiniMax Music 3. September 12, 2026 comparison; metrics use the same 192 evaluated outputs. The recorded setup uses MiniMax's official caption rewriter. This is not a matched-compute comparison.
Context
The evaluator reports these as system-level automatic metrics, not human preference. No confidence intervals are published, and systems were not evaluated with matched compute. The three pre-existing SongBench Musicality, SongBench average, and PER values were deduplicated rather than re-added.
Source WildSongBench evaluation authors
AllMusicCaps, MiniMax Music 3 (Show measurement, test conditions, and source)
Source value
0.3608542447
Metric
AllMusicCaps text/audio alignment score
Unit
ratio
Benchmark version
WildSongBench September 12, 2026 comparison; results at dataset commit 4b0e5226cdb4ba08e1774d4cfd8d5fc87bb90c37
Category
audio
Direction
Higher is better
Sample
192
Source type
independent-evaluator
Evaluator
WildSongBench evaluation authors
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. Exact WildSongBench system row MiniMax Music 3. September 12, 2026 comparison; metrics use the same 192 evaluated outputs. The recorded setup uses MiniMax's official caption rewriter. This is not a matched-compute comparison.
Context
The evaluator reports these as system-level automatic metrics, not human preference. No confidence intervals are published, and systems were not evaluated with matched compute. The three pre-existing SongBench Musicality, SongBench average, and PER values were deduplicated rather than re-added.
Source WildSongBench evaluation authors
Q3O Control, MiniMax Music 3 (Show measurement, test conditions, and source)
Source value
4.436167205
Metric
Q3O prompt-control score (0-5 scale)
Unit
points
Benchmark version
WildSongBench September 12, 2026 comparison; results at dataset commit 4b0e5226cdb4ba08e1774d4cfd8d5fc87bb90c37
Category
audio
Direction
Higher is better
Sample
192
Source type
independent-evaluator
Evaluator
WildSongBench evaluation authors
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. Exact WildSongBench system row MiniMax Music 3. September 12, 2026 comparison; metrics use the same 192 evaluated outputs. The recorded setup uses MiniMax's official caption rewriter. This is not a matched-compute comparison.
Context
The evaluator reports these as system-level automatic metrics, not human preference. No confidence intervals are published, and systems were not evaluated with matched compute. The three pre-existing SongBench Musicality, SongBench average, and PER values were deduplicated rather than re-added.
Source WildSongBench evaluation authors
AA-Music-Instrumental (Show measurement, test conditions, and source)
Source value
1,000
Metric
blind same-prompt pairwise music preference Elo
Unit
elo
Benchmark version
v1.1
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Artificial Analysis
Status
active
Retrieved at
2026-10-23
Methodology
Independent evaluation. Blind pairwise listening uses music generated from the same prompt; Artificial Analysis's validated private evaluators choose the preferred track without knowing model identity. The instrumental board covers music without vocals.
Context
The board fixes MiniMax Music 3.0 at 1,000 as its Elo reference and shows no confidence interval. It lists 2,269 in its Samples column; the value's relationship to evaluation N is not established.
Source Artificial Analysis
AA-Music-Vocal (Show measurement, test conditions, and source)
Source value
1,000
Metric
blind same-prompt pairwise music preference Elo
Unit
elo
Benchmark version
v1.1
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Artificial Analysis
Status
active
Retrieved at
2026-10-23
Methodology
Independent evaluation. Blind pairwise listening uses music generated from the same prompt; Artificial Analysis's validated private evaluators choose the preferred track without knowing model identity. The vocal board covers music with singing or vocal elements.
Context
The board fixes MiniMax Music 3.0 at 1,000 as its Elo reference and shows no confidence interval. It lists 2,810 in its Samples column; the value's relationship to evaluation N is not established.
Source Artificial Analysis

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 54 models

As of November 2026 · Source: gradually.ai popularity index

Over time

Weekly values from October 18, 2026 to November 20, 2026

MiniMax-Music3Fish Audio S2 Pro 52Nova-3 51Phonon-2 51NVIDIA Nemotron 3.5 Transcribe 49

Method changes in the timeline, a jump at these points is no market event:

  1. 1November 1, 2026: Value only with search or usage data, 30-day attention
  2. 2November 1, 2026: Method version 4 (common scale, shrinkage)
  3. 3November 9, 2026: New topics measurement V20 (daily quorum, short forms)

What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.

All specifications

Architecture

Model class
Music generation
Source: Model class
Checkpoint
MiniMaxAI/MiniMax-Music3
Source: Checkpoint

Input and output

Input
Lyrics and a detailed music description
Source: Input
Output
Song with vocals and accompaniment
Source: 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

Access
Local weights
Source: Access
Shutdown
MiniMax API (music-3.0): retired, no date published (weights remain available)
Source: Shutdown

Further details

Length
Complete songs up to five minutes
Source: Length
Audio format
32 kHz, 16-bit stereo WAV
Source: Audio format
Components
8B global LLM, 0.6B local LLM, plus Flow Matching and Flow-VAE
Source: Components
Music API
Since 2026-08-20 the paid music API is closed to new users; existing paying users can continue using it. Free music APIs are being discontinued.
Source: Music API

Sources and data date

Show all 17 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.

TypeEvidence and data date
Research statusResearch date October 4, 2026. 7 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Additional sourceMiniMax-Music3 model card (retrieved October 3, 2026) · MiniMax-Music3 model card · Editorial description reviewed October 3, 2026
Additional sourceMiniMax-Music3 repository (retrieved October 3, 2026) · MiniMax-Music3 repository · Editorial description reviewed October 3, 2026
Additional sourceMiniMax Music 3.0 release (retrieved October 3, 2026) · MiniMax-Music3 release announcement · Editorial description reviewed October 3, 2026
Additional sourceMiniMax music API availability (retrieved October 4, 2026) · MiniMax music API availability · Editorial description reviewed October 3, 2026
Additional sourceYuE2 model card (retrieved October 4, 2026)
Additional sourceWildSongBench evaluation authors (retrieved October 17, 2026)
Additional sourceA Thousand Listeners (retrieved October 22, 2026; October 23, 2026)
Additional sourceArtificial Analysis (retrieved October 23, 2026)12
Additional sourceModel metadata (retrieved October 17, 2026)
Additional sourceModel metadata (retrieved October 23, 2026)
Additional sourceOpenRouter: Popularity index (derived), CC BY 4.0
Additional sourceVercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional sourceOllama: cumulative downloads for the popularity index
Additional sourceReplicate: cumulative runs for the popularity index
Additional sourceGoogle Trending Now: limited feed selection for Germany and the US
Additional sourceCreative Commons Attribution 4.0