MiniMax Music 3
MiniMax's open-weights song generator: full songs up to five minutes with vocals, from lyrics and a style description.
In short. MiniMax Music 3 turns lyrics and a written description of the style into a complete song of up to five minutes, with singing and instruments. It needs a strong GPU, and its license asks commercial products to show the model name.
Summary
MiniMax released Music 3 as open weights on 13 August 2026. It takes lyrics with section tags such as [Verse] and [Chorus], plus a description of the music, and writes a finished song of up to five minutes as 32 kHz stereo audio. MiniMax recommends a structured description with global metadata (genre, tempo, key), vocal details and an arrangement plan, and ships a prompt-rewriting skill that expands a short idea into that form.
Inside, an 8-billion-parameter language model initialized from Qwen3-8B plans the song's long-range structure, and a 0.6B model fills in the detail frame by frame. A 2.4B flow-matching model and a 123M-parameter decoder then turn their hidden states into audio. The repository is about 57 GB. The diffusers route in the card fits a 24 GB GPU in BF16 and, with layer-by-layer offloading, even an 8 GB card, only more slowly, while the GitHub guide serves the model with SGLang-Omni on two CUDA GPUs. Inference needs CUDA, only non-streaming generation works, and a prompt is limited to 5,000 tokens and 9,000 audio frames.
MiniMax publishes no benchmark numbers for it. In the YuE2 team's own WildSongBench, MiniMax Music 3 scores 6.28 on the SongBench average, behind YuE2 (6.73) and slightly ahead of HeartMuLa (6.25), and its 6.27% phoneme error rate is the lowest of the open models in that table. The license is the MiniMax-Music3 Community License: commercial products must display "MiniMax-Music3", companies earning over US$20 million a year need written authorization, and anyone hosting the model for others must keep safeguards against infringing outputs. In the 30 days to 11 November 2026, the official repository had about 5,600 downloads, while the ComfyUI team's repacked copy had about 1 million. Download counts may include automated pulls.
Good for
- Full songs with vocals from your own lyrics on your own hardware
- Control over genre, tempo, key, vocals and section-by-section arrangement
- Developers who want a ComfyUI, diffusers or SGLang-Omni route
- Products that can show "MiniMax-Music3" in the interface
Watch out for
- The card's standard route needs a CUDA GPU with about 24 GB, and MiniMax's GitHub guide asks for two GPUs when serving with SGLang-Omni
- Section tags and descriptions steer the result but do not guarantee it, so tempo, key and lyrics may drift
- Commercial products must show the model name, and companies over US$20 million a year need MiniMax's written authorization
- The card does not describe the training data, so ask a lawyer about copyright risk before you sell the output
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Memory figures for each variant of MiniMax Music 3. Image, video, speech and music pipelines also load a text encoder and a decoder, so the memory needed can be larger than the main checkpoint.
| Variants | Parameters | Download size | Memory needed | Fits on |
|---|---|---|---|---|
| Music 3 (diffusers or SGLang-Omni) BF16 | 11.1B | 57 GB | 24 GB | Gaming GPU |
Download size covers every component needed to run the model, such as the generator, text encoder and decoder. Memory needed is a stated or published measured minimum where one exists; otherwise it is the download size plus about 15%. Larger images, longer clips and higher resolutions need more. The developer states a minimum of 24 GB of GPU memory.
The license in plain words
An MIT-style license with commercial conditions. You may use, modify and distribute MiniMax Music 3, but a commercial product must show "MiniMax-Music3" in its interface, and a business earning over US$20 million a year from products that use it needs written authorization. Anyone who offers the model to others must also keep safeguards against infringing outputs. A long acceptable-use policy applies.
- Attribution or model name must be shown
- Free commercial use only below a revenue cap
- Acceptable-use policy applies
- Outputs must be filtered or reviewed, and AI generation disclosed
More about MiniMax Music3 Community
See also
- ACE-Step 1.5 vs MiniMax Music 3: which open song model?
- Best open-source music and sound-generation models in 2026
- Open music models grew up in 2026: seven models for songs, instruments and sound effects
Frequently asked questions
- What is MiniMax Music 3?
- MiniMax Music 3 turns lyrics and a written description of the style into a complete song of up to five minutes, with singing and instruments. It needs a strong GPU, and its license asks commercial products to show the model name.
- Can I use MiniMax Music 3 commercially?
- Yes, with conditions. MiniMax Music 3 is released under the MiniMax Music3 Community, which allows commercial use but adds extra rules: Attribution or model name must be shown; Free commercial use only below a revenue cap; Acceptable-use policy applies; Outputs must be filtered or reviewed, and AI generation disclosed. Read the license text before you ship a product.
- Is MiniMax Music 3 open source?
- MiniMax Music 3 is open-weight, not open source in the strict sense. You can download the model files, but the MiniMax Music3 Community adds conditions that the Open Source Initiative definition does not allow.
- Can I run MiniMax Music 3 on my own computer?
- Yes, on a good graphics card. The smallest variant, Music 3 (diffusers or SGLang-Omni), needs about 24 GB of memory at 4-bit precision, which suits a 24 to 32 GB gaming GPU or a Mac with 32 GB or more.
- When was MiniMax Music 3 released?
- MiniMax Music 3 was released by MiniMax on August 13, 2026.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- MiniMax-Music3 model card (Hugging Face) huggingface.co
- MiniMax Music 3.0: Next-Generation Open-Weights, Production-Ready & Versatile Music Model (MiniMax, 13 Aug 2026) minimax.io
- MiniMax-Music3 repository, serving guide and limitations (GitHub) github.com
- MiniMax-Music3 Community License (Hugging Face) huggingface.co
- YuE2-3B model card, WildSongBench comparison table (Hugging Face) huggingface.co
Release date and facts last verified November 11, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| YuE2 M-A-P | 3.63B | – | CC BY-NC (Non-commercial) | 24 GB Gaming GPU | |
| ACE-Step 1.5 ACE-Step | 2B – 4B | – | MIT (Permissive) | 4 GB Laptop or phone | |
| HeartMuLa HeartMuLa | 3.94B | – | Apache 2.0 (Permissive) | 26 GB Gaming GPU | |
| Stable Audio 3 Stability AI | 460M – 1.4B | – | Stability Community (Conditional) | 4.02 GB Laptop or phone |