MiniMax Music: Lyrics-Controlled AI Songs at $0.15 a Track, Explained (Aug 2026)
Last updated August 7, 2026
MiniMax Music is the song line from the Hailuo video company: you supply exact lyrics (1-3,500 characters) with [Verse]/[Chorus] structure tags and it composes around them, up to ~5 minutes at 44.1 kHz for a flat $0.15 per generation — the market's price floor. Music-3.0 (Jul 2026) is the flagship; commercial terms are thin in the API docs. Runs in invideo's Music tab.
Updated August 2026
MiniMax Music is the song-generation line from MiniMax, the Shanghai AI lab better known for its Hailuo video models. Its pitch is control and price: you hand the model explicit lyrics — up to 3,500 characters, with section tags like [Verse] and [Chorus] marking the structure — and it composes the song around them for a flat $0.15 per generation of up to five minutes (official pay-as-you-go pricing, August 2026). That is the aggressive floor of the paid music-model market, and it comes with a cover/remix pipeline and free-tier models on top. The line has moved fast — five releases between June 2025 and July 2026 — landing on Music-3.0 as the current flagship. Here is the whole family: timeline, controls, specs, pricing, and the parts of the licensing story that are still thin.
What is the MiniMax Music timeline?
From MiniMax's official model release notes (platform.minimax.io/docs/release-notes/models.md):
| Model | Date | What changed |
|---|---|---|
| Music-1.5 (beta Jun 20, 2025; full Sep 11, 2025) | 2025 | "Music from musical prompts and lyrics"; full release added 4-minute songs |
| Music-2.0 | Oct 29, 2025 | "Enhanced musical expression that makes every melody deeply emotional" |
| Music-2.5 | Jan 16, 2026 | Quality iteration |
| Music-2.5+ | Mar 2026 | "Instrumental Unlocked" — instrumental-only generation |
| Music-2.6 | Apr 2026 | "Cover Reborn. Bass Redefined." — rebuilt cover mode |
| Music-3.0 | Jul 16, 2026 | Current flagship |
MiniMax positions Music-3.0 around better semantic understanding — it "reduces the usual AI-generated drift," in the docs' phrasing — plus "commercial-grade audio quality" with instrument-level techniques like slides and legato, and vocal synthesis with control over "melody, pronunciation, breathing, and harmonies" (music generation guide).
How much control do you get over the song?
More explicit control than most music models, because the lyrics are an input, not a suggestion:
- Lyrics are required for vocal tracks — 1 to 3,500 characters, your exact words. Structure tags
[Intro],[Verse],[Pre-Chorus],[Chorus],[Bridge],[Hook], and[Outro]tell the model how to arrange the song, so you dictate where the chorus hits rather than hoping. A Lyrics Generation API ($0.01 per song) can draft the words if you don't have them. - Instrumental toggle. The
is_instrumentalflag (introduced with Music-2.5+) produces vocal-free tracks from the same prompt pipeline. - Cover/remix pipeline. The dedicated
music-covermodel takes reference audio from 6 seconds to 6 minutes and reworks it, extracting the original lyrics automatically via ASR — the mode Music-2.6's "Cover Reborn" release rebuilt.
MiniMax Music specs and limits
| Spec | Detail |
|---|---|
| Current models | music-3.0, music-2.6, music-cover (paid, 120 RPM) + -free variants at 3 RPM |
| Lyrics input | 1–3,500 characters with section tags; required for vocals |
| Duration | Up to ~5 minutes (the pricing unit); no hard maximum documented |
| Audio quality | Sample rates 16/24/32/44.1 kHz; bitrates up to 256 kbps |
| Formats | MP3, WAV, PCM, delivered by URL |
| Cover mode | Reference audio 6 s–6 min, ASR lyric extraction |
Source: MiniMax music generation guide and API reference, as of August 2026.
The sleeper row is the -free variants: 3 requests per minute is a throttle, but a genuinely $0 API tier is something almost no other current music vendor offers.
How much does MiniMax Music cost?
$0.15 per generation for up to 5 minutes of music, for both music-3.0 and music-2.6, per MiniMax's official pricing page (August 2026). Lyrics generation costs $0.01 per song, and the -free model variants cost nothing at a 3-requests-per-minute cap. No other frontier music vendor publishes a lower per-track price for five-minute output as of this writing — pricing is the family's clearest competitive position, especially for creators generating many candidates and keeping one.
Can you use MiniMax music commercially?
This is the family's weakest documentation, and it deserves a straight answer: the API docs publish no commercial-use terms at all as of August 2026. MiniMax's consumer-facing music pages assert that users own their outputs and can use them commercially, but that language lives outside the API documentation — verify it against the platform's current terms of service before relying on it for client or broadcast work. There are no licensed-training claims, no disclosed rightsholder deals, and no watermarking or provenance disclosure. The practical read: unlike vendors that at least publish a rights grant alongside their pricing, MiniMax's developer paper trail is silent — whatever rights you hold live in consumer-page marketing language and the current terms of service, not in the API docs you would actually build against.
Straight answers on MiniMax Music
What is the difference between Music 3.0 and Music 2.6?
Music-3.0 (July 16, 2026) is the flagship quality tier — stronger semantic understanding of prompts, more realistic instrument articulation, finer vocal control. Music-2.6 (April 2026) remains available at the same $0.15 price with the rebuilt cover mode as its headline. Both take the same lyrics-and-tags input.
Do I have to write my own lyrics?
For vocal tracks, yes — lyrics (1–3,500 characters) are a required input, which is precisely what makes the output controllable. If you want help, the Lyrics Generation API drafts them for $0.01 per song, and instrumental mode skips lyrics entirely.
How long can a MiniMax song be?
The pricing unit covers generations up to about 5 minutes; MiniMax does not document a hard maximum beyond that. Music-1.5's full release was marketed on 4-minute songs, so ~4–5 minutes is the realistic working range.
Can MiniMax remix an existing song?
The music-cover model reworks reference audio between 6 seconds and 6 minutes long, extracting the source lyrics automatically via ASR. Rights in the source material are yours to clear — the model does not check them.
Is there a free way to use the MiniMax Music API?
Yes — free variants of the music models run at 3 requests per minute, one of the few no-cost API tiers among current music generators.
Does MiniMax make other models besides music?
Yes — MiniMax is the company behind the Hailuo video family and a Speech/voice line, both of which also run inside invideo. The MiniMax hub on invideo covers the full stable in one place.
Where can you use MiniMax Music?
Minimax Music 3.0 and Music 2.0 are both selectable in the Music tab of the invideo agent, alongside the rest of the platform's 200+ models — so the same conversation that writes and generates your song can also cut a video to it with MiniMax's own Hailuo models or any other engine on invideo. Since the model's superpower is exact-lyrics input, the natural pairings are the lyric video maker — you already have the words, timed and structured — and the music video maker for building visuals around the finished track.
Version history: Music-1.5 beta (Jun 20, 2025) → Music-1.5 full, 4-minute songs (Sep 11, 2025) → Music-2.0 (Oct 29, 2025) → Music-2.5 (Jan 16, 2026) → Music-2.5+ instrumental (Mar 2026) → Music-2.6 cover rebuild (Apr 2026) → Music-3.0 (Jul 16, 2026).