Skip to content
All news

AI music

Mureka V9.5 vs V9: what the music benchmarks show

Mureka V9.5 leads V9 in a vocal-song preference test. Their instrumental scores are nearly identical. Here is what artists can infer from the results.

DeadMod News

Published

Sources checked / 3 min read

AA-Music v1.1 leaderboard snapshot checked October 9, 2026

The facts

Who
Artificial Analysis, its recruited listeners, and music generators including Mureka.
What
Separate preference rankings for songs with vocals and instrumental music.
Where
The public Artificial Analysis Music Arena leaderboards.
When
The v1.1 results and methodology were checked October 9, 2026.
Why
A model's usefulness depends on the kind of music an artist needs.
How
Listeners compare generated tracks without seeing the model names.

Mureka V9.5 scores higher than V9 in Artificial Analysis's test of songs with vocals. Their instrumental results are nearly identical. For an artist choosing music for a video, those are two different reasons to audition a model.

Where does Mureka V9.5 lead V9?

The vocal leaderboard, checked October 9, 2026, puts V9.5 at 1103 and V9 at 1052. That is a 51-point gap within this benchmark. It is not a 51 percent improvement.

Test Mureka V9 Mureka V9.5
Songs with vocals 1052 1103
Instrumental music 1081 1080

The vocal board lists 95 percent confidence intervals of 1038 to 1066 for V9 and 1088 to 1118 for V9.5. These ranges express uncertainty around the estimated ratings. They do not overlap in this snapshot.

Why does the instrumental result need a different reading?

The instrumental leaderboard, checked October 9, shows a one-point difference between the models. Both share a listed rank range of third through sixth. Their confidence intervals overlap substantially.

Our analysis: This result does not establish a useful instrumental advantage for either version. It also does not prove that every instrumental they produce will sound equally good. The vocal result alone is not evidence for replacing V9 in an instrumental workflow.

Does the benchmark measure accurate lyrics or reliable edits?

The methodology, checked October 9, 2026, describes blind comparisons by recruited listeners. Public Arena votes do not affect the ratings. Higher scores reflect listener preference within each test.

For the vocal test, the models write their own lyrics from prompts. The benchmark does not supply a fixed lyric sheet. An attractive result therefore does not show whether a model will sing your exact words correctly.

The ratings also do not establish edit accuracy, download reliability, or permission to release the result. Those require separate checks. Artificial Analysis says providers do not pay for listings or favorable results. We reviewed its published results and did not reproduce the listening study.

Impact on users

Our analysis: Use the relevant board to choose candidates for a small audition. A sung demo and an instrumental bed for a release teaser have different requirements. A higher score cannot choose the right performance for your project.

For a lyric video, listen for missing words, pronunciation, and changes in the final chorus. Fix the audio before you build detailed lyric timing. A caption cannot repair a different word in the recording.

If you compare versions, record the attempts needed to obtain a usable file. Include those attempts in your budget. This report establishes no subscription saving or price advantage for either model.

What needs a check before release?

Keep the chosen audio file, project settings, and relevant permissions together. Confirm collaborator consent and credit. Follow the music video release checklist before delivery.

The figures here describe the October 9 leaderboard snapshot. A later model, different genre, or specific lyric brief can produce a different outcome.

Sources and reporting

This article uses the public sources below. AI assisted the research and draft. It includes no interviews or hands-on tests. Sections marked “Our analysis” explain possible effects on users.

  1. Artificial Analysis: AA-Music-Vocal v1.1 leaderboard
    Checked October 9, 2026
  2. Artificial Analysis: AA-Music-Instrumental v1.1 leaderboard
    Checked October 9, 2026
  3. Artificial Analysis: music benchmark methodology
    Checked October 9, 2026

DeadMod also makes music software. Read our editorial standards and correction policy.