ElevenLabs Music v2.5 became the default: which model made a track?

ElevenLabs made Music v2.5 the default on 2026-09-11. When a default changes, old and new tracks differ. How to record the engine per track, on Sume too.

4 min readSume
All posts

If you call a music API without naming a model, the engine behind your tracks can change under you: ElevenLabs made Music v2.5 its default on 2026-09-11, with API support following on 2026-09-14, per a changelog roundup. Tracks made the week before and the week after may come from different models. The fix is the same for every vendor: pin the model where the API lets you, and store the engine that actually ran with each track.

Sume's Music Router works this way: omit model and Sume picks, or pass an explicit id and keep control.

Nothing in the ElevenLabs roundup says the change was announced as breaking; it is simply a new default. That is the normal way a model change reaches you, which is why the logging habit matters more than any one vendor's notice period.

How do Sume's ids behave?

The request's model is optional. Omitted, it means sume/music-auto, which resolves to Lyria 3.5 today. The other routable ids are lyria-3.5 and lyria-3-pro. The job echoes the id you sent in job.model, and job.request.routed_model names the engine that ran, for example lyria-3.5.

Sume Music Router ids from the Music Router docs, with ElevenLabs' reported change, read 2026-10-02.
Id you sendWhat runsWhat the job records
(omitted) or sume/music-autoLyria 3.5 todayjob.model = sume/music-auto; routed_model = the engine
lyria-3.5Lyria 3.5job.model = lyria-3.5
lyria-3-proLyria 3 Projob.model = lyria-3-pro
ElevenLabs default (reported)Music v2.5 since 2026-09-11See the vendor's output metadata

Why does an auto id matter for a brand sound?

An auto id is a promise that something good runs, not that the same thing runs next month. If a campaign needs one consistent score, pin an explicit id and re-test before changing it. If you only need a usable bed, auto is fine, as long as you log routed_model so you can tell tracks apart later.

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: pinned-score-001" \
  -d '{"model":"lyria-3.5",
       "prompt":"Warm lo-fi, 84 BPM, C minor. Dusty Rhodes, brushed drums. A 30-second track. Instrumental, no vocals."}'

What should I log per track?

Four things: the id you requested, the engine that ran, the date and the job id. Add the vendor's own version string for any tool that reports one. Store these four fields in the same row as the downloaded file, not in a separate note that gets lost. When a client says the new music sounds different, you can answer from the log instead of guessing. The Music Router page documents the fields, and Jobs and results shows how to read a result.

Sources

Related posts

More in Developers

All Developers posts

Written by Sume