"Instruments sound live, and tracks stay coherent from start to finish."
ElevenLabs has released its new music model, Music v2.5.
Hi everyone, this is Sonetho. ⚡
On September 11, 2026, ElevenLabs announced the release of Music v2.5.
This update comes about three and a half months after Music v2 launched in late May.
This post covers only the verified facts from the official announcement and documentation.
We haven't included our own hands-on audio comparisons here. We'll share our ears-on impressions in a separate post once we test it ourselves.
🎧 What's Changed
1. Thicker, Fuller Sound
According to the official description, instruments sound closer to live performances, tracks have more layers and movement, and that quality holds up from start to finish.
They also specifically noted an increase in overall arrangement depth.
2. Outperformed v2 in Blind Tests
ElevenLabs conducted 47,885 blind pairwise comparisons and stated that v2.5 was chosen in the majority of matchups.
However, they did not disclose the exact percentage, so we won't speculate on specific numbers.
3. Major Improvements in Specific Genres
The announcement highlighted the biggest improvements in vocal-driven and acoustic-heavy genres.
R&B, soul, hip-hop, rock, metal, orchestral, and cinematic were specifically named.
Conversely, the difference may be less noticeable in electronic, synth-heavy tracks.
💾 Lossless Downloads Now Available on Free Plans
This is the most notable change in this update.
Free Plan: 5 lossless downloads per day
Pro Plan: 400 lossless downloads per month
However, downloads are disabled for tracks created using reference audio from other artists.
Keep this in mind when using the reference feature to avoid unexpected hurdles.
📍 Where You Can Use It
Music v2.5 is available across all of the following:
ElevenMusic (Dedicated music interface)
ElevenCreative (All-in-one subscription)
Studio (Adding music to video)
Flows (Node-based connections)
API
The existing Music v2 is not going away. In official terms, it's "not going anywhere."
If you're already used to how v2 sounds, you can continue using it as is.
⚙️ One Important Detail for API Users
While you don't need to worry about this in the web interface, the API is a different story.
The API reference states that the default model_id is music_v1.
The selectable values are music_v1, music_v2, and music_v2_5.
Interestingly, in the model list within that same official documentation, ElevenLabs explicitly labels music_v1 as "outclassed" by v2 and v2.5.
In other words, if you make an API call without specifying a model, it runs on a model the creators themselves consider outdated.
If you plan to use v2.5 via the API, make sure to set model_id to music_v2_5. It's just a one-line difference.
Track length ranges from 3 seconds to 5 minutes, and reference audio clips can be up to about 30 seconds.
Outputs are provided in MP3 (44.1 kHz) and WAV formats.
💼 Commercial Use
ElevenLabs states that generated tracks can be used for most commercial purposes, including videos, advertisements, podcasts, and games.
However, tracks created on the Free plan require attribution.
The required phrasing in the terms is "Created in collaboration with ElevenLabs,"
though other approved variations may also be used.
There is also a welcome clause:
Even if you downgrade or cancel your plan, you retain the rights to keep using tracks created prior to the change.
On the other hand, the guidelines clearly list what cannot be included in prompts:
Real or stage names of singers and composers (living or deceased)
Song titles, album titles, and record label or publisher names
Substantial portions of lyrics from existing songs
Generating outputs that imitate the voice or signature style of a specific artist is also prohibited.
Because commercial terms vary depending on the use case (such as streaming distribution), it's best to check the rights table in the terms of service if you plan to release your tracks publicly.
✍️ Summary
To sum up this update in one sentence:
The sound is fuller, and lossless downloads are now available even on the free tier.
If you create music featuring vocals or acoustic instruments, it's well worth running your existing prompts through v2.5.
If your tracks are mostly electronic, the difference might be less pronounced.
Just like we did with v2, we plan to compare v2 and v2.5 directly using the same lyrics.
We'll share what differences we can hear with our own ears, and where they stand out, in our next post.
Thanks for reading! ⚡