Project AURIONAI Use-Cases Research & Innovation
MiniMax
music and audio

MiniMax Music 3

MiniMax open-weight music model, up to 5 min at 32 kHz stereo; top open-weights model and 1000 anchor on AA Music Arena v1.1. Licence file not read.

MiniMaxMiniMaxChina
MiniMax Music 3 architecture diagram from the model card.
Levelmodel
Classmusic and audio
Parametersnot in the notes
Contextnot in the notes
Released7 Aug 2026
Opennessunknown
Runs onsingle datacentre GPU
Statusactive

Benchmark Results

independent
Artificial Analysis Music Arena v1.1generative media
secondary capture of AA post; anchor; top open weightsindependent
1,000
8001,0001,200

Score in Elo; the axis starts at 800, as Elo has no zero. Higher is better.

rank 13, AA Music Arena v1.1, InstrumentalOct 2026 alphasignal.ai, digg.com
secondary capture of AA post; anchor; top open weightsindependent
1,000
8001,0001,200

Score in Elo; the axis starts at 800, as Elo has no zero. Higher is better.

rank 11, AA Music Arena v1.1, VocalOct 2026 alphasignal.ai, digg.com
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).

Lineage

finetunecontinued pretrain
Qwen3-8B to Molmo2-8B: finetune, high confidenceQwen3-8B to OneRec-Foundation 8B: continued pretrain, high confidenceQwen3-8B to MiniMax Music 3: continued pretrain, medium confidenceQwen3-8BQwen3-8BAlibaba (Qwen), Apr 2025Molmo2-8BMolmo2-8BAllen AI, Dec 2025OneRec-Foundation 8BOneRec-Foundation 8BKuaishou, Jan 2026MiniMax Music 3MiniMax Music 3MiniMax, Aug 2026
Alibaba (Qwen)Qwen3-8B is the parent (continued pretrain)medium confidencemodel card
Where this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 4 models. Confidence and evidence are the notes' own.

R&D Topics

Text-to-music and song generationproduced by

Identity

Developer
MiniMax
Family
not in the notes
Released
7 Aug 2026 sourcesregistry date
Weights published
–
Record checked
10 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.

Architecture and Size

Architecture
hybrid Hierarchical AR Hybrid-LM (Global 8B from Qwen3-8B plus Local 0.6B) plus flow matching plus Flow-VAE
Total parameters
– sources
Active parameters
–
Checkpoint parameters
–
Note
About 11B across components
Context length
not in the notes
Input
text
Output
audio
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).

Openness and Access

Openness
unknown
Licence
not in the notes sources
Access routes
open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.

Deployment

Runs on
single datacentre GPU
Footprint
CUDA required
Runtimes
not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.

Indonesia

Indonesian support
not stated
Regional languages
not in the notes
Jakarta hosting
unknown
Languages
not in the notes
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.

Industry Use Cases

No industry use case links this model.

Sources

  1. MiniMaxAI/MiniMax-Music3 , Hugging Face huggingface.co https://huggingface.co/MiniMaxAI/MiniMax-Music3
  2. Text-to-Audio Models – Hugging Face huggingface.co https://huggingface.co/models?pipeline_tag=text-to-audio&sort=trending
  3. Artificial Analysis' Music Arena v1.1 Crowns Suno v6 Across 1,000 Harder Prompts | AlphaSignal alphasignal.ai https://alphasignal.ai/news/artificial-analysis-music-arena-v1-1-crowns-suno-v6-across-1-000-harder-prompts

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "MiniMax Music 3", https://aurion.semai.id/models/minimax-music-3/, records of October 2026.