August 13, 2026Model Release

MiniMax-Music3

MiniMax

MiniMax put a production-grade music model on Hugging Face, weights included. Music3 pairs an 8B language model for song structure with a 0.6B model for acoustic detail, feeding a flow-matching decoder that renders 32 kHz stereo WAV up to five minutes.

The timing is not subtle. A week earlier Suno announced download caps, seven lifetime downloads for free users, and for anyone scoring a film the download is the product. The license is the interesting part: unlike MiniMax-H3 eleven days earlier, there is no territorial carve-out. Free worldwide, with mandatory "MiniMax-Music3" branding in commercial products and written authorization above $20M revenue.

32 kHz 16-bit stereo WAV · Up to 5 minutes · 8B global LLM + 0.6B local LLM + 2.4B flow matching

Part of 2026 · The Shakeout

Last updated October 2, 2026