Chinese AI company MiniMax released Music 2.6 on April 10, rebuilding the model's generation architecture to cut initial latency to under 20 seconds. The update focuses on two things: a new Cover function and tighter integration with the AI agent ecosystem.
Cover is the headline addition. Users upload an existing track, and the model extracts its melodic structure, letting them swap out genre, arrangement, or instrumentation while keeping the original motif and rhythm intact. MiniMax is pitching it as a re-creation tool rather than a simple remix engine, though how well it preserves musical identity across radical genre shifts remains to be tested independently.
Alongside the model, MiniMax open-sourced three MusicSkill modules: minimax-music-gen2 for music creation, minimax-music-playlist for personalized playlist generation, and buddy-sings, which gives virtual assistants a singing voice. These plug into MiniMax's broader agent ecosystem and are available for developers to integrate into third-party services.
The company also claims improved mixing quality and better prompt adherence for parameters like BPM, key, and instrument entry order. Low and mid-frequency response got specific optimization, per coverage from AIBase. Music 2.6 is available in beta with a 14-day free trial.
Bottom Line
MiniMax Music 2.6 drops generation latency under 20 seconds and adds a Cover function for reshaping existing tracks, with a 14-day free beta now open.
Quick Facts
- Release date: April 10, 2026 (UTC+8)
- Initial generation latency: under 20 seconds (company-reported)
- Three open-sourced MusicSkills: minimax-music-gen2, minimax-music-playlist, buddy-sings
- 14-day free beta trial available
- Listed company: MINIMAX-W (00100.HK)
