Artificial intelligence has become increasingly capable of writing songs, and a relatively unfamiliar competitor has now beaten Suno, arguably the biggest name in AI music, at its own game.

ACE Music v1, developed by ACE Studio, has taken first place on Artificial Analysis' vocal music leaderboard, scoring 1,160 Elo points against 1,144 for Suno v6.

The AA-Music-Vocal v1.1 benchmark measures the quality of AI-generated songs containing vocals. Evaluators listen to tracks produced from identical prompts without knowing which model created them, then choose their favorite. Those preferences determine an Elo-style rating, with higher scores indicating stronger listener preference. Because Elo-style ratings are relative rather than absolute, the scores reflect head-to-head preference within the pool of models tested rather than an independent measure of quality.

The lead is relatively narrow, and the two models' statistical confidence intervals overlap, meaning the ranking gap can fall within the range of measurement uncertainty. Suno also retains the top position on the separate instrumental music leaderboard.

ACE Studio, founded in 2018, originally focused on synthesizing singing voices from MIDI notes and lyrics — the same vocal territory the new leaderboard measures. Its latest flagship model generates complete songs from written descriptions or user-provided lyrics.

Decrypt's hands-on tests: stronger melodies, cleaner endings

In Decrypt's hands-on tests, ACE Music consistently produced more melodic and coherent songs in mainstream styles. Its vocal performances generally sounded more natural, with fewer of the synthetic artifacts that sometimes make AI singers sound as though they are struggling to pronounce their own lyrics, relying heavily on autotune, or hallucinating words simply to fill empty space.

Decrypt even published one of its own test tracks — a song about the outlet's latest company announcement regarding the Information Exchange — to demonstrate the model's output.

ACE also handled song endings better. Where Suno occasionally introduces strange sounds, unnecessary repetitions, hallucinations, or abrupt musical changes near the finish, ACE generally maintained its structure and delivered cleaner conclusions. That matters because producing a convincing chorus is relatively easy compared with maintaining a musical idea throughout an entire composition, even after the lyrics have already ended.

Suno still held an advantage when the models were pushed toward less conventional genres. In Decrypt's melodic death metal test, ACE consistently produced tracks — one example here — featuring a woman singing clearly in a pleasant, clean voice, missing the aggressive vocal delivery expected from the requested style. More broadly, the model confused metal genres across the board. Suno offered more creative arrangements in the proper heavy styles, with better stylistic consistency across several similarly demanding prompts.

Those results were subjective rather than a controlled benchmark, but they showed why a leaderboard victory does not automatically make one model the better choice for every musician.

More than a song generator

ACE's most interesting feature may be its songwriting workflow. Its “Idea” feature can develop a basic concept into lyrics and a structured composition, identifying individual song sections and describing how they should be arranged. Users can then refine the material rather than repeatedly generating completely new tracks.

The v1.1 Preview goes further, creating a musical score before generating audio. “Each generation is a new performance, so the sound may differ even when the score stays the same,” the developers explain.

Users can edit melodies and chords, regenerate performances, export MIDI files, and separate vocals from instruments for further production. MIDI itself is a long-standing standard format for encoding notes and performance data, so exported files can be opened in conventional music production software for further editing. That workflow places ACE closer to a music production assistant than a simple prompt-to-song application.

Open-source ambitions

The flagship ACE Music v1 model remains closed-source with no public API, so users can currently generate music only through the official site. Its developers are nonetheless pursuing a parallel open-source strategy. ACE Studio and StepFun have already released ACE-Step 1.5, a separate family of models whose code and weights are publicly available. Unlike the proprietary flagship, these models can run on users' own hardware — though at the cost of weaker performance.

The company has also promised another release. “We plan to open-source ACE Music v1 Flash in late 2026,” the developers state on their website, describing a version designed to balance quality, generation speed, and hardware requirements.

ACE Music is available through its browser application, with a limited free tier and paid subscriptions. Commercial usage rights for flagship-model songs require a paid plan — a distinction that matters for anyone planning to release or monetize generated tracks — while ACE-Step offers a separate free generation option.

For now, ACE Music v1 holds a 16-point Elo lead over Suno v6 in vocal generation. Its promised Flash release would give users running models locally another option to try before the end of 2026.