ACE Studio's ACE Music v1 Beats Suno to Top AI Music Leaderboard

ACE Music v1, from Chinese AI music company ACE Studio, has debuted at number one on Artificial Analysis’s AA-Music-Vocal v1.1 leaderboard. Its Elo score places it ahead of Suno v6, Suno v6-mini, and Mureka V9.5 in the benchmark’s current snapshot.

The hosted model generates complete songs, vocals, or instrumentals from a text description or supplied lyrics. Its arrival puts a vocal-synthesis specialist ahead of established text-to-music providers on a public blind-listening benchmark.

ACE takes a narrow lead

AA-Music-Vocal v1.1 calculates Elo ratings from blind pairwise votes. Evaluators hear two clips generated from the same prompt, choose the stronger result, and remain unaware of which model produced either clip. Elo measures relative preference within that pool, while production suitability still depends on genre, prompt, workflow, and licensing requirements.

ACE Music v1 leads Suno v6 by 16 points and Suno v6-mini by 28. Reported confidence intervals are approximately plus or minus 20 points, so the gap over Suno v6 remains within the benchmark’s uncertainty range and may shift as more votes arrive.

Vocal synthesis grows into full songs

ACE Studio is an AI music workstation developed by Timedomain Inc. The product opened a closed beta in April 2022, entered public beta that July, and released version 1.8.0 in October 2023.

The workstation began with tools for producing and editing synthetic vocal performances. Producers can control pitch, timing, expression, MIDI-based performances, and AI instruments at a finer level than one-prompt song generators typically provide. ACE Music v1 extends that work into end-to-end song generation.

The team also co-developed the open-weights ACE-Step repository with StepFun. According to the project, ACE-Step v1.5 can run on a consumer GPU, generate a full song in under two seconds on an Nvidia A100, and support LoRA fine-tuning. Those speed and quality claims come from the developers and require independent testing. ACE Music v1 remains a separate, closed model delivered through ACE Studio’s hosted service.

Reggae and orchestral prompts lead

Artificial Analysis found the model strongest on reggae, dancehall, classical, and orchestral prompts. Published examples include a 1970s roots-reggae track with nyabinghi hand drumming and a comic-opera patter aria featuring bassoons and audience laughter. These tasks reward stable instrumentation, intelligible vocals, and sustained musical character.

Results appeared less consistent on dense modern pop production and deliberately awkward combinations such as math rock with Midwest emo vocals. One test requested a restless singer performing slightly off-key, a form of controlled imperfection that remains difficult for current music generators.

Hosted app, no public API

ACE Music v1 was available through the ACE Music app at publication, with no public API for automated generation or pipeline integration. The listed access options were:

• Free tier: 20 credits per day
• Paid plans: starting at $7.90 per month when billed annually
• Highest tier: up to 500 songs per month

ACE-Step provides the local, open-weights option for teams that need self-hosting or fine-tuning. Its capabilities, model behavior, and output quality should be evaluated separately from the hosted ACE Music v1 service.

Listeners can sample the model without creating an ACE account through the Music Arena. The arena presents anonymous pairs generated from identical prompts, and those votes feed the leaderboard’s Elo ratings.

From ranking to production test

ACE Music v1’s result suggests that expertise in controllable singing voices can transfer to complete song generation, particularly in vocal-forward genres. The related ACE-Step project also gives researchers and self-hosters an open model from the same broader development effort, although its benchmark results should not be treated as interchangeable with those of ACE Music v1.

Teams evaluating generated vocals, demos, or scratch tracks should run identical prompts and lyrics through ACE Music v1, Suno v6, and Mureka V9.5, normalize playback levels, and compare the outputs blindly. Useful evaluation criteria include:

• Lyric intelligibility and pronunciation
• Vocal consistency across sections
• Song structure and prompt adherence
• Instrument separation and production artifacts
• Generation latency, failure rates, and revision controls
• Commercial-use rights, output ownership, and data-retention terms

The leaderboard gives ACE Music v1 the highest average preference score in this test, while the narrow lead and substantial genre variation leave model selection dependent on each project’s material and operational requirements.