ACE Music v1 tops Artificial Analysis vocal leaderboard, narrowly edging Suno v6
In brief
- ACE Music v1 scored 1,160 Elo on Artificial Analysis' vocal leaderboard, ahead of Suno v6's 1,144.
- Confidence intervals for the two models overlap; Suno still leads the instrumental leaderboard.
- Decrypt's subjective tests favored ACE in mainstream styles and Suno in heavier, niche genres.
- ACE Music v1 is closed-source with no public API; developers plan to open-source v1 Flash in late 2026.
How the ranking works
The AA-Music-Vocal v1.1 benchmark measures the quality of AI-generated songs that contain vocals. Evaluators listened to tracks built from identical prompts without knowing which model made them, then picked a favorite. Those picks set an Elo-style rating.
That's a 16-point gap. Decrypt noted that the lead is relatively narrow and that the confidence intervals overlap, so it isn't a runaway result.
ACE Studio was founded in 2018 and originally focused on synthesizing singing voices from MIDI notes and lyrics. Its ACE Music v1 model generates complete songs from written descriptions or from lyrics that users supply.
What Decrypt's tests found
Decrypt also ran its own hands-on comparison, which it described as subjective rather than a controlled benchmark. In mainstream styles, ACE consistently produced more melodic and coherent songs, and its vocals generally sounded more natural, with fewer synthetic artifacts. The outlet said ACE also handled song endings better than Suno, keeping its structure and delivering cleaner conclusions.
Niche genres were a different story.
Suno performed better once Decrypt pushed both models toward less conventional styles. In a melodic death metal test, ACE kept producing clean female vocals instead of the aggressive delivery the genre calls for, while Suno offered more creative arrangements in heavy styles and better stylistic consistency across several similarly demanding prompts.
Tools, access and open-source plans
ACE's "Idea" feature develops a basic concept into lyrics and a structured composition (it identifies song sections and describes how to arrange them). The v1.1 Preview goes further by creating a musical score before generating audio. Users can edit melodies and chords, regenerate performances, export MIDI files and separate vocals from instruments.
There's a catch for developers, though. ACE Music v1 remains closed-source with no public API, so songs can only be generated through the official site. ACE Studio and StepFun did release ACE-Step 1.5, a separate model family with publicly available code and weights; it can run on users' own hardware but is weaker than the proprietary flagship.
“We plan to open-source ACE Music v1 Flash in late 2026,” the developers state
For now, it's a stated plan from ACE's developers rather than a shipped product.
Frequently asked questions
How does Artificial Analysis rank AI music models?
The AA-Music-Vocal v1.1 benchmark measures the quality of AI-generated songs with vocals. Evaluators listen to tracks made from identical prompts without knowing which model created them, then choose a favorite. Those preferences set an Elo-style rating.
Is ACE Music v1 clearly better than Suno?
Not clearly. ACE Music v1 scored 1,160 Elo to Suno v6's 1,144 on the vocal leaderboard, but the lead is narrow and the confidence intervals overlap. Suno still leads Artificial Analysis' separate instrumental leaderboard.
Is ACE Music v1 open source?
No. ACE Music v1 is closed-source with no public API, so users can only generate songs through the official site. ACE Studio and StepFun released ACE-Step 1.5, a separate and weaker model family with public code and weights. The developers say they plan to open-source ACE Music v1 Flash in late 2026.
