Hero title card reading “ACE Music v1 Lands at #1” with the subtitle “AA-Music-Vocal v1.1: 1,160 Elo, 2,119 blind comparisons” and a five-bar vertical ranking motif on the right, the top bar accented blue and labelled “1” and “ACE Music v1”. The OrcaRouter logo is composited in the bottom-right corner.
Guides & Insights

ACE Music v1 Lands at #1 on AA-Music-Vocal — Beating Suno v6 by 16 Elo Points and One Overlapping Interval

Author

Rowan Sterling

Date Published

Latest models · 20View all models →
Benchmarks: Artificial Analysis · updated daily
Back to all posts

The headline number is 1,160. ACE Music v1 sits at the top of Artificial Analysis's AA-Music-Vocal v1.1 leaderboard with an Elo of 1,160 and a 95% interval running from 1,139 to 1,181. Second place is Suno v6 at 1,144, interval 1,126 to 1,162. Those two bands overlap across 23 Elo points, which means the honest reading of this week's result is narrower than the press line: a new model from a company most Western listeners have never heard of is now rank one on the only independent listening benchmark this category has, and the evidence that it is definitively better than Suno's flagship is suggestive rather than settled.

ACE Music v1 is the first full-song generation model from ACE Studio, the Shenzhen-founded company that has spent seven years building AI singing voices rather than AI songs. ACE Studio started in 2018 as a vocal synthesis tool — you feed it MIDI and lyrics and it sings — and only expanded into a general music workstation and then a consumer song generator. That lineage is visible in the result: the model leads a vocal benchmark and does not appear at all on the instrumental one.

What the leaderboard actually measured

AA-Music-Vocal v1.1 is not a prompt-adherence test, a mixing test, or a "did the lyrics scan" test. It is a preference benchmark. Artificial Analysis generates the same prompt across every model in the pool, strips the labels, and has validated private evaluators listen to pairs and pick a winner. Elo is computed from those votes. ACE Music v1's entry carries 2,119 comparisons — a smaller sample than any Suno or ElevenLabs entry on the board, which is what you would expect from a model that joined the pool within the past few weeks against models that have been collecting votes since spring.

The size of the win is worth putting in plain terms. Sixteen Elo points is roughly a 52–48 edge in a single blind head-to-head, not a rout. If you played a hundred pairs of ACE Music v1 and Suno v6 tracks to a listener, you would expect about fifty-two of them to go to ACE and forty-eight to Suno, with plenty of noise on either side of that. The intervals say the same thing more formally: ACE's floor of 1,139 is below Suno v6's ceiling of 1,162, so a rerun of the same comparison with fresh voters could plausibly end with the two swapped.

None of that makes the result uninteresting. A 2,119-vote debut that lands above a model with 3,063 votes is a real signal, and it is the first time since Suno v6 launched on 9 September 2026 that anything has sat above it on this board. What it does not support is the stronger claim floating around launch coverage — that ACE Music v1 is now demonstrably the best music model available. For that you would want the intervals to separate, and they do not yet.

• Rank and Elo — ACE Music v1 #1 at 1,160 (1,139–1,181) vs Suno v6 #2 at 1,144 (1,126–1,162)

• Sample size — 2,119 blind comparisons for ACE Music v1 vs 3,063 for Suno v6

• Board tenure — ACE Music v1 is a weeks-old entry with 2,119 comparisons; Suno v6 has been collecting votes since its September 2026 launch and sits at 3,063

• Third place — Suno v6-mini at 1,132, meaning Suno holds two of the top three slots

• Next tier down — Mureka V9.5 at 1,101, Mureka V9 at 1,063, then Google's Lyria 3.5 at 1,050

What ACE Studio actually shipped

The product is a web app at acemusic.ai, and it is a song generator in the same sense Suno is: you describe a style, mood and subject, and it returns a complete track with vocals. The input surface is broader than the pitch usually conveys. Text-to-song and lyrics-to-song are the defaults; the model also takes a melody played on an on-screen keyboard, a hummed recording, or an uploaded MIDI file, and builds the arrangement around your notes. There is image-to-song, an AI Cover that reworks a reference track in a different style, and a Mashup Maker that takes two reference songs as inputs.

Two model versions ship side by side. ACE Music v1 is the model on the leaderboard and the default in the free tier. ACE Music v1.1 Preview, which is a separate picker entry, writes a score in ABC notation — melody, rhythm and chords — before it generates any audio, so the composition step is inspectable and editable in a songwriting assistant before the performance step runs. The vendor's own split is clear: v1 if you are starting from a description or lyrics, v1.1 Preview if you already have a melody and want the model to develop around specific notes.

The company's other product, ACE Studio, remains the workstation: a DAW-like environment with AI vocal tracks, AI instruments, VST3 and AU plug-in support, voice cloning, and 140-plus voices across eight languages on the Verse 2.x vocal synth models. The Artist plan is the bridge between the two — it unlocks ACE Studio access and is, in effect, the plan for anyone who wants to edit what the song generator produced rather than just download it.

Generated single-column scoreboard card titled “ACE Music v1 — the scoreboard” with six rows: Vocal rank #1 of 14; Vocal Elo 1,160; 95% interval 1,139 to 1,181; Blind comparisons 2,119; Instrumental board not ranked; Public API none. Footer reads “Elo, rank and interval per Artificial Analysis, AA-Music-Vocal v1.1. No vendor-reported benchmark scores.” The OrcaRouter logo is composited in the bottom-right corner.

The vocals-only result is the story underneath the story

Artificial Analysis runs two separate music boards. AA-Music-Vocal v1.1 ranks models that generate singing; AA-Music-Instrumental v1.1 ranks models that generate music without it. ACE Music v1 is absent from the instrumental board entirely, which lists nineteen models from Suno, Mureka, Google, ElevenLabs, MiniMax, Stability, Sonauto and Meta.

That absence is not a footnote. It tells you what kind of model this is. ACE Studio's entire technical history is singing voice synthesis — the Verse family of vocal models, phoneme-level editing, breath and dynamic controls, multilingual pronunciation. Building a full-song model on top of that heritage produced something that wins on vocals and does not compete on instrumental-only work. If your job is a cue, a bed, or a score with no singer, this model is not aimed at you and the leaderboard result does not transfer.

It also explains the shape of the top of the board. Suno v6 is the only model in the pool that sits near the top of both lists — first on instrumentals at 1,148, second on vocals at 1,144. ACE Music v1 has one strong number and one blank. That is a specialist's profile, and it is worth more to a singer-songwriter than to a sync library.

What it costs, in credits and in dollars

ACE Music prices in credits, and the conversion is unusually easy to work with: one generation produces two songs and costs four credits, so two credits buys one song. Uploading audio, recording a hum and playing the keyboard are free.

• Free — $0, 20 credits a day (up to 10 songs), shared queue, no downloads and no commercial rights

• Creator — $7.90/month on annual billing, 1,000 credits (about 500 songs), unlimited downloads, stem splitter, commercial use rights, no ACE Studio access

• Artist — $16.58/month on annual billing, 2,500 credits (about 1,250 songs), everything in Creator plus ACE Studio, voice cloning, plug-in support and the music video creator

At the Creator rate a song costs roughly one and a half cents, which is far below what the same volume costs on any subscription music platform priced per track. The catch sits on the free tier: no downloads at all, so the free plan is an audition rather than a workflow. There is also a permanently free open-weights sibling, ACE-Step v1.5, co-developed with StepFun, which the vendor offers with unlimited generations, free MP3 downloads and commercial use — a 4-billion-parameter diffusion transformer in its XL variant, and a different model from the one that topped the leaderboard.

The missing piece is the one developers will notice first

There is no public developer API for ACE Music v1 that we can find as of 11 November 2026. The documentation site covers the ACE Studio application, not an inference endpoint; the obvious API paths on the product domain return 404; no platform subdomain resolves. Every model in this comparison that competes on access — Eleven Music, Lyria 3 Pro — has a published endpoint and a published per-song price. ACE Music v1 currently has a web app, a credit balance and a download button.

For a solo writer that is fine, and possibly preferable: the pricing is simpler than a metered API and the editing tools come bundled. For a team that wants music generation inside a product, the absence changes the decision more than sixteen Elo points do. You cannot fail over to a model you cannot call, which is exactly the gap a routing layer exists to close — OrcaRouter puts 200-plus models behind one endpoint with provider list price passed through at 0% markup and automatic failover between providers, and the reason that matters here is not that music models are on it today (they are not) but that the moment a vendor publishes a listable music endpoint, the integration is a model string and a fallback rather than a second contract. Until then, ACE Music v1 is a destination, not a dependency.

What would make this result durable

Three things are worth watching over the next quarter, and none of them is another launch post.

The first is whether the interval narrows. ACE Music v1 has 2,119 votes; the models around it have three thousand or more. If the model keeps collecting comparisons at the rate a #1 ranking attracts and the point estimate holds near 1,160, the lower bound climbs past Suno v6's upper bound and the argument ends. If the estimate drifts down toward 1,145 as the sample grows, this was a debut effect and the board was right to keep Suno v6 within reach.

The second is instrumentals. ACE Studio has the vocal half of the problem solved in a way few companies do. Whether it ships an instrumental-competitive model — or whether it stays a vocal specialist and leaves cues and beds to Suno, Lyria and Stable Audio — determines whether this is a niche winner or a category leader.

The third is access. A model that tops an independent benchmark and cannot be called programmatically is a marketing asset more than an infrastructure one. If an API appears with a per-song price, the comparison against Lyria 3 Pro's $0.08 and Eleven Music's credit pricing becomes possible; until it does, the price comparison in this article is a subscription comparison, and that is a different question.

For anyone choosing right now: if you write songs with a singer in them and you are tired of re-rolling until the vocal stops sounding synthetic, ACE Music v1 is worth the seven dollars and ninety cents it costs to test properly, downloads included. If you need instrumental beds, you are shopping on the wrong board. And if a number in a launch post told you this model is the best music AI on earth, read the interval before you repeat it.

Screenshot of the Artificial Analysis music leaderboard in English, AA-Music-Vocal, showing the ranked table with ACE Music v1 in first place at 1,160 Elo and the Suno v6 row immediately below it at 1,144, with each row's confidence interval and comparison count.Screenshot of the ACE Music homepage at acemusic.ai in English, showing the “#1 Music Model / ACE Music v1” badge and an AI MUSIC TOOLS grid of generation tools above the prompt field.