99.9% SLA
MiniMax Music 2.6: Full-Length Song Generation with Cover & Music Skill - Full Review
MiniMax Music 2.6 (this catalog entry: minimax-music-2-6, API minimax-music-v2.6) is MiniMax's next-generation music generation model, released April 10, 2026. It generates full-length songs up to 6 minutes, auto-writes lyrics, and adds two things the music AI category rarely ships together: a Cover creation feature and a Music Skill for AI agents.
Here's the short version: Music 2.6 is the most complete music offering MiniMax has shipped - longer songs (up to 6 minutes, most landing between 2 and 4), improved latency and acoustic quality, and direction over genre/mood/scene through text. The Cover feature is the standout: create song covers from existing tracks, which opens the remix/cover workflow inside the same model. The honest caveats: launched with a 14-day free beta window (500 free generations/day for consumers, 100 API calls/day for developers), the quality claims are vendor-reported, and Cover workflows carry licensing obligations you own.
This guide covers model overview, core features, technical specifications, capability comparison, core advantages, recommended use cases, example prompts, selection recommendations, workflow, and my verdict.
Quick Facts
| Attribute | Value |
|---|---|
| Catalog slug | minimax-music-2-6 |
| API model | minimax-music-v2.6 |
| Developer | MiniMax |
| Release | April 10, 2026 |
| Song length | Up to 6 minutes (most 2-4 min) |
| Lyrics | Auto-generated or custom (with structure tags) |
| Headline features | Cover creation, Music Skill (agents) |
| Free beta | 14 days; 500 gens/day consumer, 100 API calls/day dev |
| Access | MiniMax platform + API |
| Self-hosting | No |
Table of Contents
- Model Overview
- Core Features
- Technical Specifications
- Capability Comparison
- Core Advantages
- Recommended Use Cases
- Example Prompts
- Selection Recommendations
- Workflow: Song, Cover, Agent
- The Bottom Line
- FAQ
- Sources & Further Reading
1. Model Overview
MiniMax Music 2.6 is the April 2026 refresh of MiniMax's music generation line. The release notes focus on three improvements over earlier versions: generation latency, music control, and acoustic quality - plus the two new capabilities that make it a platform rather than a generator: Cover and Music Skill.
The model generates complete songs or instrumentals from text descriptions (style, mood, scene), supports custom lyrics with structure tags, and auto-writes lyrics when you do not provide them. Duration reaches 6 minutes per generation, with most songs landing between 2 and 4 minutes - full-song length, not clips.
The Cover feature creates cover versions from existing tracks, and Music Skill exposes generation to AI agents - MiniMax's bet that music becomes an agent-callable capability, not just a consumer toy.
2. Core Features
Full-length songs. Up to 6 minutes per generation; most songs land between 2 and 4 minutes.
Auto-generated lyrics. Describe the theme; the model writes the lyrics.
Custom lyrics with structure. Provide lyrics with [verse]/[chorus]/[bridge] tags for exact control.
Text-to-music direction. Genre, mood, scene, and style from natural language.
Cover creation. Generate cover versions from existing tracks - the remix workflow.
Music Skill for AI agents. Agent-callable music generation for the AI ecosystem.
Improved latency and acoustics. The three headline quality upgrades over prior versions.
Free beta window. 14 days of generous free access for testing.
3. Technical Specifications
| Specification | Detail |
|---|---|
| Model | Music 2.6 (minimax-music-v2.6) |
| Max duration | 6 minutes (most 2-4 min) |
| Lyrics | Auto or custom (structure tags) |
| Tasks | Full song, instrumental, cover |
| New features | Cover, Music Skill |
| Release | April 10, 2026 |
| Free beta | 14 days (500/day consumer, 100 calls/day dev) |
| Access | MiniMax platform + API |
Note: The 14-day free beta ran from launch (April 2026). Current quotas and pricing may differ - verify the live console.
4. Capability Comparison
| Capability | MiniMax Music 2.6 | Lyria 3 Pro (Google) | ACE-Step (open) | HeartMuLa (open) |
|---|---|---|---|---|
| Max length | 6 min | Full songs | ~4 min | Full songs |
| Auto lyrics | Yes | Yes | Tags/lyrics | Yes (lyrics-first) |
| Cover feature | Yes | No | No | No |
| Agent skill | Yes (Music Skill) | No | No | No |
| Open source | No | No | Yes | Yes |
| Managed API | Yes | Yes (preview) | No | No |
Reading the table honestly: Music 2.6's differentiators are Cover and Music Skill - neither open family has them, and Lyria 3 Pro does not advertise Cover. For ownership, ACE-Step/HeartMuLa remain the open picks.
5. Core Advantages
- Cover creation. The remix/cover workflow inside the same model - unique in the managed tier.
- Music Skill for agents. Music becomes callable by AI agents - the forward-looking feature.
- Full-song length. Up to 6 minutes; no clip stitching.
- Auto + custom lyrics. Both paths, with structure tags for control.
- Quality refresh. Latency, control, and acoustic quality all improved.
- Generous free beta. Real testing headroom at launch.
6. Recommended Use Cases
- Songwriting: full songs from prompts or your own lyrics.
- Covers and remixes: Cover feature for existing tracks (with licensing diligence).
- Agent-powered music: Music Skill for AI agents that generate on request.
- Content music: style/mood/scene-driven tracks for videos and games.
- Instrumental production: song-free instrumentals from text direction.
7. Example Prompts
1. Full song with auto lyrics
2. Custom lyrics
3. Cover request
Prompting guidance:
- State duration, genre, BPM, and vocal direction first - the model composes from structure.
- Use structure tags for custom lyrics; alignment is a core strength.
- For covers, name the target style and tempo explicitly; licensing is your responsibility.
8. Selection Recommendations
Choose MiniMax Music 2.6 if:
- Cover creation is part of your workflow.
- You want agent-callable music (Music Skill).
- Full-song length with auto or custom lyrics on a managed API fits your stack.
Choose Lyria 3 Pro if:
- You are in the Google ecosystem and want DeepMind's managed full-song model.
Choose ACE-Step / HeartMuLa if:
- Open source, local control, and no per-song fees matter more.
Test before committing if:
- Acoustic quality is brand-critical - launch claims need your ears.
9. Workflow: Song, Cover, Agent
Practical notes:
- Keep a style vocabulary per brand for consistent music direction.
- For covers, confirm rights before monetizing.
- Use the agent skill for pipeline integration; direct API for interactive work.
10. The Bottom Line
Verdict: Buy for cover workflows and agent-ready music - the most platform-minded music model in the managed tier. MiniMax Music 2.6 delivers full-song length, auto lyrics, and solid quality refresh, but the differentiators are Cover and Music Skill - capabilities that move music generation from a toy endpoint to a creative platform. The open families (ACE-Step, HeartMuLa) remain the ownership picks, and Lyria 3 Pro is the Google-ecosystem alternative. For managed full-song generation with cover and agent paths, Music 2.6 is the strongest current option.
Sources & Further Reading
- MiniMax Music 2.6 release - ITHome
- MiniMax Music 2.6 launch - China News Shanghai
- MiniMax music API via TokenHub - Tencent Cloud
- Music 2.6 model page - Replicate (hosting partner)
Information reflects MiniMax's official release and hosting-partner documentation as of August 2026. Quotas and pricing evolve; verify the live console.
