A 3-minute AI music track: ElevenLabs $0.45 vs Sume $0.125, 3.6x

ElevenLabs lists music at $0.15 per minute, which makes a 3-minute track $0.45. Sume Music charges a flat $0.125 per generation, so the same length costs 3.6 times less ($0.45 / $0.125 = 3.6). The catch is that Sume has no duration parameter. You ask for the length in the prompt, and the model decides what comes back.

Prices were read on ElevenLabs' API pricing page and Sume's catalog on 2026-10-09. Sume's docs describe the output as full-length structured songs of up to a few minutes.

Price against the length you actually get

Because the length is guided by the prompt, price the risky case. If you ask for three minutes and get 2:20, you paid $0.125 for 2:20, which is still cheaper than ElevenLabs at that length ($0.35). If you get 1:30, ElevenLabs would have charged $0.225 for it. The flat fee only loses when the track is under 50 seconds.

How to ask for length on Sume

The docs suggest writing the length into the brief, such as a 2-minute track, or using section markers like [0:00-0:30] Intro. Do not send duration or duration_seconds, which the API rejects, and do not send a non-empty negative_prompt, which returns 400 with negative_prompt_unsupported. Put exclusions in the positive prompt, for example Instrumental, no vocals. The prompt can be up to 5,000 characters.

The result is a Sume-hosted audio artifact, usually audio/mpeg on media.sume.com. Read the duration from the file, not from your prompt.

When to prefer which

If your project needs a track of an exact length, such as 30.0 seconds for an ad slot, a flat fee with no length control means you may trim or regenerate. Each regeneration is another $0.125. Per-minute pricing that follows the length requested is easier to budget for that case.

For long background beds where a minute either way does not matter, Sume's flat fee is hard to beat on these prices.

• Long beds and songs: Sume's flat fee wins by a wide margin.
• Exact-length stings: expect trimming or retries.
• Check the delivered length before it goes into a timeline.

A ten-track example

Suppose you need ten background beds of about three minutes each for a set of videos. At ElevenLabs' per-minute price that is 10 x $0.45 = $4.50. On Sume it is 10 x $0.125 = $1.25 if every one lands near three minutes, and $2.50 if you need two attempts for each.

Even at two attempts per track, the flat fee is $2.00 cheaper than a single pass at the per-minute price, so on long tracks you can afford to regenerate. That changes how you work: write a precise brief, listen to the first result, and only then decide whether to try again.

Whatever you pick, check the licence terms for commercial use on the vendor's own pages before you publish music in an ad.