MiniMax Music API closed to new users: what Sume offers instead
MiniMax's pay-as-you-go page says its paid Music API is closed to new users from August 20, 2026. Sume's Music Router is open: one fixed price per track.

If you wanted to start calling MiniMax's music API today, the vendor's own pricing page says you cannot: starting August 20, 2026, its paid Music Generation and Lyrics Generation APIs are no longer available to new users, while existing paying users can keep using them. Sume's Music Router has no such cutoff in its docs: you send a prompt and get one track for a fixed $0.125.
What does MiniMax say on its pricing page?
The Music section of the MiniMax pay-as-you-go page carries the notice quoted above and points people to MiniMax Audio (the consumer product) or to the open-source MiniMax Music 3 model on Hugging Face instead. Every music row on the page is labelled Discontinued. The page was read on 2026-10-03.
| Model | Listed price | Rate limit on page |
|---|---|---|
| Music-3.0 (Discontinued) | $0.15 per up-to-5 minutes of music | RPM = 120, contact sales to increase |
| Music-2.6 (Discontinued) | $0.15 per up-to-5 minutes of music | RPM = 120, contact sales to increase |
| Music-2.0 (Discontinued) | $0.03 per up-to-5 minutes of music | not stated |
| Lyrics Generation (Discontinued) | $0.01 per song | not stated |
What does the Sume Music Router do?
Music Router is Sume's music surface. Omit model or set it to sume/music-auto and Sume picks the engine (Lyria 3.5 today, per the docs). You can also pin a catalog id: the routable ids listed are sume/music-auto, lyria-3.5 and lyria-3-pro. The request body is a prompt of 1 to 5000 characters plus an optional image_url.
Pricing is stated in the docs as one fixed Music price per audio generation, and the Music 1.0 page gives it: $0.125 per accepted generation, with no change for prompt length or image conditioning. The catalog lists each engine's provider list price for reference, but the charge is the fixed price.
How do the two compare side by side?
The table is about access and billing shape, not audio quality, which neither page measures.
| Question | MiniMax pay-as-you-go page | Sume Music Router docs |
|---|---|---|
| Can a new account use the paid music API? | No, from August 20, 2026 | Yes: submit to POST /v1/music-router/generate |
| What is billed? | $0.15 per up-to-5 minutes of music (Music-3.0, Music-2.6) | $0.125 per accepted generation, fixed |
| How is length set? | Priced per up-to-5 minutes | duration and duration_seconds are rejected; steer length in the prompt |
| Engine choice | One model id per row | sume/music-auto or a pinned catalog id |
What changes in your code?
Because the Music Router rejects duration fields, any code that passed a length to another music API must move that intent into the prompt text, for example "a 2-minute track" or a timed section map such as [0:00-0:30] Intro. The docs also say exclusions belong in the positive prompt, because a non-empty negative_prompt is unsupported.
A minimal call looks like this:
curl -X POST https://api.sume.com/v1/music-router/generate \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: music-router-001" \
-d '{"model": "sume/music-auto", "prompt": "Warm lo-fi hip hop, 84 BPM. A 30-second track. Instrumental, no vocals."}'When is the open-weights route the better fit?
MiniMax's own page names the open-source Music 3 model as the alternative for people who cannot get API access. That route means you host and operate the model yourself, which is a different cost and effort profile from a per-track API price. If you only need finished tracks for a video, a hosted router is the shorter path; if you need to modify the model, the weights are the point.
Either way, plan for the time you will spend: the Music Router returns a job, so you poll status or use a webhook, then read the audio artifact from result.artifacts[] where type is audio. The result can also carry result.lyrics with the model-reported lyrics or section map when the engine provides one.
Is the price comparison like for like?
Not exactly, and the arithmetic should be read with that in mind. MiniMax priced up to five minutes of music per request, so a short jingle and a five-minute song cost the same on its page. Sume prices each accepted generation at a flat $0.125 and does not let you pass a duration, so the length you get is whatever the engine produces for your prompt. Twenty tracks at the Sume rate come to $2.50 by simple multiplication, but twenty tracks of unknown length are not twenty five-minute songs. Generate a few and listen before you budget a campaign around either number.
Sources
Related posts
More in Comparisons
- MiniMax speech-to-text $0.38 per hour vs Sume STT $0.01 per minute
MiniMax lists speech-to-text at $0.38 per hour. Sume's STT is $0.01 per audio minute ($0.60 per hour), reached through video inspect on a hosted clip.
- Mubert Render cost per generated track: Creator $0.028, Pro $0.078
Mubert Creator is $14 for 500 tracks a month ($0.028 each, non-commercial) and Pro $39 for 500 with commercial use ($0.078). Sume Music is $0.125.
- Music cost: ElevenLabs $0.15/min vs Sume Music 1.0 $0.125/track
ElevenLabs lists music at $0.15 per minute; Sume Music 1.0 is a flat $0.125 per generation. Break-even is 50 seconds; a 2 minute track is $0.30 vs $0.125.
- Notta Pro 1,800 minutes, 5-hour files vs Sume STT 10-minute jobs
Notta Pro lists 1,800 monthly minutes and recordings up to 5 hours at $8.17 a month billed annually. Sume STT is $0.01 a minute in 600-second jobs.
Written by Sume