MiniMax Music 3.0

Spec sheet

Purpose

MiniMax Music 3 generates complete songs from a creative concept and optional lyrics in a single generation.Read 12 Oct 2026 · minimax.io

Supported platforms

webRead 22 Sep 2026 · minimax.io

Free plan

YesRead 22 Sep 2026 · minimax.io

Support

The Music 3 model card lists [email protected] as its contact address.Read 12 Oct 2026 · huggingface.co

License

The MiniMax-Music3 Community License requires commercial products using the software to prominently display “MiniMax-Music3” and requires prior written authorization when aggregate yearly revenue exceeds US$20 million.Read 12 Oct 2026 · huggingface.co

Founded

2022Read 28 Sep 2026 · minimax.io

Company

MiniMax describes itself as a global AI foundation model company founded in early 2022.Read 12 Oct 2026 · minimax.io

Audio export formats

wavRead 22 Sep 2026 · minimax.io

Audio output

The model produces 32 kHz, 16-bit stereo WAV audio.Read 12 Oct 2026 · huggingface.co

Availability

MiniMax says its music models stopped being available through Token Plan and M Plan on August 20, 2026, and directs users to MiniMax Audio or the open-source model weights.Read 12 Oct 2026 · platform.minimax.io

Commercial use

YesRead 22 Sep 2026 · minimax.io

Company products

MiniMax says its models power products including MiniMax Code, MiniMax Design, MiniMax Audio, Talkie, and its enterprise and developer-facing Open API Platform.Read 12 Oct 2026 · minimax.io

Control caveat

Section tags and music descriptions guide generation but do not guarantee that tempo, key, instruments, lyrics, or structure will exactly match the request.Read 12 Oct 2026 · huggingface.co

Generation limits

The model card lists a 5,000-token text prompt limit, a 9,000-acoustic-frame audio limit, and non-streaming generation only.Read 12 Oct 2026 · huggingface.co

Hardware requirement

Inference requires CUDA; the model card says automatic CPU offloading can reduce GPU memory needs to about 22 GB, with slower layer-by-layer offloading fitting 8 GB cards.Read 12 Oct 2026 · huggingface.co

Local use

The official model repository documents downloading the model and running it with Diffusers, including a device-map example using CUDA.Read 12 Oct 2026 · huggingface.co

Model access

MiniMax publishes downloadable model weights and instructions for local use with SGLang, Diffusers, and ComfyUI.Read 12 Oct 2026 · huggingface.co

Model components

The model uses an 8B Global LLM for long-range structure, a 0.6B Local LLM for acoustic detail, and a synthesis stack with flow matching and Flow-VAE.Read 12 Oct 2026 · minimax.io

Music controls

Its inputs can specify lyrics and musical details such as genre, tempo, key, emotion, vocal delivery, instrumentation, arrangement, and production profile.Read 12 Oct 2026 · huggingface.co

Open weights

MiniMax describes Music 3.0 as an open-weights model, and its official Hugging Face repository provides model files for download.Read 12 Oct 2026 · huggingface.co

Prompt enhancement

A template-based Prompt Enhancement System can expand a simple description into detailed musical instructions.Read 12 Oct 2026 · minimax.io

Song length

The model is designed to sustain creative intent across songs of up to five minutes.Read 12 Oct 2026 · minimax.io

Song structure

Lyrics can use section tags such as intro, verse, pre-chorus, chorus, bridge, instrumental, solo, and outro to define macrostructure.Read 12 Oct 2026 · minimax.io

Structured captions

Structured Captions can specify genre, tempo, time signature, key, use case, production character, emotional contour, instrumentation changes, groove, and vocal delivery.Read 12 Oct 2026 · minimax.io

Vocal controls

Structured Captions can describe vocal timbre, delivery, breathiness, falsetto, harmony arrangement, and effects such as delay and Auto-Tune.Read 12 Oct 2026 · minimax.io

Vocal generation

YesRead 22 Sep 2026 · minimax.io

Where Specifiction places it

• AI music generators#104 of 1237.3