MiniMax Music 2.6

• 🎵 Generates full-length songs with vocals and instrumentation from lyrics
• 🆕 Adds instrumental-only mode, Lyrics Optimizer, and Cover transformation
• 📚 Rich structure tags in lyrics shape arrangement and song sections
• 🎯 Cover mode reimagines an existing track while preserving its melody
• 🔧 Lyrics Optimizer auto-writes lyrics from your style prompt
• 💬 Designed to allow deliberately imperfect vocals for lo-fi and indie
• 🌐 Ships with open-sourced "Music Skills" for AI agent developers
• 🏢 Built by Chinese AI lab MiniMax

MiniMax

MiniMax is an AI company building generative models across multiple modalities, with a focus that spans both language understanding and audio creation. Their rapid release cadence in early 2026—delivering several new models within just a few months—reflects…

MiniMax Music 2.6 is the latest entry in MiniMax's song-generation family, producing complete, full-length tracks with vocals and instrumentation from a text style prompt and lyrics, or instrumental-only output from a prompt alone. Per the provider's documentation, users describe genre, mood, vocal style, and instruments, then shape the song with rich structure tags inside the lyrics to mark sections such as the intro, verse, chorus, bridge, and outro, with synchronous or streaming delivery through the Music API.

Compared with its same-family predecessor MiniMax Music 2.5, MiniMax describes version 2.6 introducing three notable additions: an instrumental-only mode, a Lyrics Optimizer that auto-writes lyrics from your prompt, and an AI Cover feature. With Cover, you upload a track, the model extracts its melodic skeleton, and you can switch genre, instrumentation, and lyrics while keeping the recognizable melody. MiniMax also says 2.6 lets vocals and melody be deliberately imperfect, suiting lo-fi, indie folk, and indie jazz.

The lineage traces back through MiniMax Music 2.0. Alongside the 2.6 release, MiniMax open-sourced three "Music Skills" aimed at AI agent developers, letting agents identify intent, select original, instrumental, or Cover mode, and call the generation API automatically.

MiniMax also develops separate text and speech families, including MiniMax M3 and MiniMax Speech-02 HD.