MiniMax Launches Music 2.6, Open-Sources Three Agent Skills

MiniMax has released Music 2.6, the newest version of its AI music generation model, and open-sourced three "MusicSkills" designed to plug music creation directly into the growing ecosystem of AI agents. The release lands at a moment when AI music tools are moving from standalone web apps to building blocks that autonomous agents can call on demand.

The headline feature is AI Cover mode, which transforms an existing song into a new style while preserving its original melody. Alongside that, a new Lyrics Optimizer can generate full lyrics from a short prompt, meaning users no longer need to bring their own words to the model.

What's New in Music 2.6

Music 2.6 generates a complete, full-length track — vocals, layered instrumentation, and genre-aware mixing — in seconds from a text prompt. Previous versions already handled text-to-song, but 2.6 adds two capabilities that broaden the range of use cases.

AI Cover mode takes an input song and re-renders it in a different genre or style. You could take a folk ballad and get a synthwave cover, or a pop chorus and get a gospel arrangement, with the underlying melody intact. This is a meaningful step beyond simple style transfer because it preserves musical identity while swapping out everything else.

Lyrics Optimizer closes the loop for users who have an idea but not the words. Give it a prompt like "a breakup song about outgrowing a city," and it returns a full lyric sheet, ready to be handed back to the model for generation. For creators who can describe a feeling but freeze up when writing verses, this removes a major friction point.

Every track produced by Music 2.6 is labeled as royalty-free AI music and cleared for commercial use from the moment it is generated — a detail that matters for anyone building media products on top of these APIs.

The Open-Source MusicSkills

The more strategically interesting part of the announcement is the three open-sourced MusicSkills, designed to be called by AI agents rather than end users.

The first, minimax-music-gen2, is aimed at musicians and handles high-quality track generation from natural-language prompts. The second, minimax-music-playlist, produces personalized playlists based on user context, listening history, or mood descriptions. The third, buddy-sings, is a "pet singer" skill that generates playful vocal tracks — more of a consumer fun case than a pro tool.

Why this matters: AI agents are starting to need access to creative primitives the same way they need access to databases or calendars. If an agent is helping plan a birthday party, it is increasingly reasonable to expect it to generate a custom song as part of the workflow. Open-sourcing these skills, rather than locking them behind a proprietary API, is a bet that music generation will become a standard capability in the agent stack.

Industry Impact

Music generation has been one of the most competitive corners of generative AI. Suno, Udio, Stability Audio, and a growing list of Chinese labs have all pushed model quality forward over the past year. MiniMax's bet with 2.6 and its skills release is less about beating those competitors on raw audio quality and more about winning distribution through the agent ecosystem.

For agent developers, a free, open-source music skill that plugs into frameworks like MCP is a much lower barrier to experimentation than signing up for another paid API. If the skill works well, it gets embedded in agent templates and tutorials, and that becomes a distribution channel in itself.

For rights holders, AI Cover mode raises familiar questions. MiniMax's own outputs are labeled royalty-free, but cover functionality depends on using input songs that may themselves be copyrighted. The company has not publicly detailed how it polices or licenses the input side of covers, and that is likely to be a flashpoint as the feature gets more use.

Expert Perspective

Coverage from AIbase and other outlets framed Music 2.6 as the entry point into an "AI music cover era," where remixing and re-arranging existing tracks becomes as accessible as creating new ones. That framing tracks with how the feature is being demonstrated: short, shareable reinterpretations of well-known songs.

Developers on X reacted most strongly to the skills release, noting that the playlist and generation skills slot neatly into existing MCP-style agent setups. The availability of an official MiniMax MCP server — which already covers text-to-speech, image, and video generation — makes it straightforward to extend an agent with music capabilities in a few lines of config.

What's Next

Music 2.6 is available now via the MiniMax platform and API, and the three open-source MusicSkills are published for integration into agent frameworks. MiniMax has signaled that this is an ongoing track of work, with more skills and higher-fidelity models expected through the rest of 2026.

For creators, the short-term move is to try AI Cover mode on their own catalog or public-domain tracks and see how the feature holds up on different genres. For developers, the interesting experiment is wiring minimax-music-gen2 into an existing agent and seeing what new product ideas fall out when music generation is essentially free to call.

Bottom Line

MiniMax is not trying to win AI music by being the single app where everyone goes to make songs. It is trying to make music a default capability of AI agents everywhere, and Music 2.6 plus the open-sourced skills are the building blocks for that strategy. For anyone building agents in 2026, adding music generation just got meaningfully cheaper and easier.