Try AnyToSpeech
Music models · updated 2026-11-22 · test set music-v2

Music Model Benchmark

The same 24 instrumental briefs and 12 songs with lyrics given to every music generator. Scores are measured from the audio: words sung, beats per minute, seconds of length.

8models tested
36tracks per model
288tracks requested
6scored tests
Lyric accuracyGenreTempoLengthNo vocals when askedClean endingSpeed
Rank 1
Stable Audio 2.5
Stability AI · instrumental only
93.5 / 100
Lyricsn/a Genre87.5 Tempo91.7 Length100 No vocals100 Ending95.8
Rank 2
Eleven Music
ElevenLabs
91.6 / 100
Lyrics97.1 Genre87.5 Tempo75 Length100 No vocals100 Ending100
Rank 3
ACE-Step 1.5
ACE Studio / StepFun
88.2 / 100
Lyrics82.8 Genre66.7 Tempo100 Length100 No vocals100 Ending100

Leaders

The top model on the measures people ask about most.

Overall ranking

Score out of 100, higher is better. Weighted: Lyrics 30%, Genre 20%, Tempo 20%, Length 10%, No vocals 10%, Ending 10%.

0255075100

Instrumental-only models are scored on the tests they support, so their overall score leaves out lyrics. Speed and reliability are reported separately and do not affect this score.

Every model on every test

Darker is better. Click a column heading to sort by that test, or a test name to open its page.

Model Overall LyricsGenreTempoLengthNo vocalsEnding Speed
1 Stable Audio 2.5Stability AI 93.5 n/a 87.5 91.7 100 100 95.8 3 s
2 Eleven MusicElevenLabs 91.6 97.1 87.5 75 100 100 100 5 s
3 ACE-Step 1.5ACE Studio / StepFun 88.2 82.8 66.7 100 100 100 100 8 s
4 Lyria 3 ClipGoogle 85.2 98 83.3 91.7 100 100 8.3 14 s
5 Lyria 3.5Google 79.8 99.2 83.3 66.7 0 100 100 23 s
6 Lyria 3 ProGoogle 77.4 98.3 83.3 58.3 0 95.8 100 26 s
7 MusicGen Large (stereo)Meta 74.4 n/a 75 79.2 100 100 12.5 73 s
8 MiniMax Music 2.6MiniMax 64.6 99.6 70.8 12.5 0 100 80.6 135 s
50 or lower100Hover a cell for its range and sample size.

Accuracy against speed

Up is more accurate, left is faster. Hover a point for its name.

Called directly (full request time) Through Replicate (compute time only)

The two kinds of point are timed differently, so compare speed within a kind, not across.

Compare models

Pick up to three. Each spoke is one test, scored 0 to 100.

What stands out

82.8+

Lyrics are sung accurately

The lowest lyric accuracy among the 6 models that sing. Misses are mostly tracks that ran out of time.

3 of 8

Length requests ignored

Lyria 3.5, Lyria 3 Pro, MiniMax Music 2.6 almost never returned the 30 seconds asked for.

12.5 to 100

Tempo is the hardest instruction

The spread on hitting a requested BPM, from MiniMax Music 2.6 to ACE-Step 1.5.

2

Well-known names not tested yet

Suno (no public api), Udio (no public api).

The tests

Each has its own page with the ranking, the rule, and audio.

Think a score is wrong? Check it.

Every test item, scoring rule and known limit is published, and the audio behind each number is on the page.

Read the methodology