ACE Music v1 vs ElevenLabs Music: A Vocal Champion Against the Only Music Model With a Developer API

One of these models tops the independent listening chart and has no API. The other sits 144 Elo points below the leader on that same chart and is the most straightforward music model in the world to call from code. ACE Music v1 is ACE Studio's first full-song generator, ranked first on Artificial Analysis's AA-Music-Vocal v1.1 at 1,160. ElevenLabs Music — currently shipped as Eleven Music v2.5 — ranks eighth on the same board at 1,016, and is available through a documented API with a published per-minute overage rate. Which one you want depends on a question that has nothing to do with taste: do you need to hear the song, or do you need to ship it inside something?

The gap on vocals is real but should be read with the same care as any new leaderboard entry. ACE Music v1's 1,160 comes from 2,119 blind comparisons; Eleven Music v2.5's 1,016 comes from 3,248. A 144-point Elo gap is a genuine separation — roughly a 70/30 edge in a head-to-head — and unlike the ACE-versus-Suno margin, these two intervals do not overlap. On singing, the newer model from the smaller company is ahead by a margin the statistics support.

Where each one actually sits on the board

ElevenLabs has three music models in the pool, and their ordering is not the ordering you would guess from the version numbers.

• ACE Music v1 — #1 on vocals at 1,160 (1,139–1,181), 2,119 comparisons, absent from the instrumental board

• Eleven Music v2.5 — #8 on vocals at 1,016, #11 on instrumentals at 1,001, released 11 September 2026

• Eleven Music v2 — #9 on vocals at 1,002, #9 on instrumentals at 1,011, released 27 May 2026

• Eleven Music (the original) — #11 on vocals at 998, #7 on instrumentals at 1,021, released 5 August 2025

Eleven Music v2.5 improved on its predecessor for singing and slipped slightly for instrumentals. The original 2025 model is still the best ElevenLabs entry on the instrumental board. That is not unusual in a fast-iterating category — a model tuned for one kind of output drifts on another — but it does mean "newest ElevenLabs music model" and "best ElevenLabs music model for your task" are two different sentences, and the pricing page will not tell you that.

The bigger structural point is that ElevenLabs is competitive on both boards while ACE Music v1 appears on only one. Suno v6 and ElevenLabs both field models that can carry an instrumental cue; ACE Music v1 cannot be evaluated on that task at all because it is not measured there. If your work is scoring a video, ElevenLabs is the safer bet almost regardless of the vocal number.

The price comparison is where the two stop being comparable

ACE Music sells a subscription with a generous song allowance. ElevenLabs sells credits that clear per generation and per audio minute, with a published overage rate that makes the real cost legible but higher.

• ACE Music Creator — $7.90/month billed yearly, 1,000 credits, about 500 songs, unlimited downloads and commercial rights

• ACE Music Artist — $16.58/month billed yearly, 2,500 credits, about 1,250 songs, plus the full ACE Studio workstation

• ElevenLabs Starter — $6/month (first month $1), 30,000 credits, roughly 8–16 tracks, extra minutes at about $0.18

• ElevenLabs Creator — $22/month, 121,000 credits, roughly 34–67 tracks, extra minutes at about $0.16

• ElevenLabs Pro — $99/month, 600,000 credits, roughly 167–333 tracks, extra minutes at about $0.15

Do the division and the shape of the difference appears. ACE Music's Creator plan works out to under two cents a song. ElevenLabs' Creator plan, at the midpoint of its own estimate, lands somewhere between thirty and sixty cents a track — an order of magnitude apart, for models that are 144 Elo apart on vocals in ACE's favour and roughly level on instrumentals.

That comparison is not quite fair, and it is worth saying why. ElevenLabs' credits are shared across its entire platform — speech synthesis, voice cloning, dubbing, sound effects and music all draw from the same balance, so the music line item is cheaper than the plan price suggests if you are already paying for the rest. The company also publishes an API price, which ACE Music does not, and a published API price is worth real money to a team that needs to forecast spend on a per-call basis rather than a per-seat one.

The licensing fine print decides more deals than the leaderboard

Both services grant commercial rights on paid plans, but the boundaries are drawn differently and the difference matters if you are producing for anyone other than yourself.

ElevenLabs' self-serve tiers — Starter, Creator, Pro — permit commercial use with an explicit carve-out for film, television and studio games; full online and offline commercial rights sit behind the Enterprise tier. Release to streaming services such as Spotify and Apple Music is not offered on self-serve plans. The free tier is personal use only. For a podcaster or an ad agency making social cuts, the self-serve terms are enough. For a production house scoring a broadcast spot, they are not, and the path is a sales conversation rather than a checkout page.

ACE Music's Creator tier grants commercial use rights without those category exclusions as published, and downloads are unlimited on both paid tiers. The free plan has no commercial rights at all and, more importantly, no downloads — you can listen to what you made but you cannot take it. There is one exception worth knowing: the separate open-weights model ACE-Step v1.5, co-developed with StepFun, is offered on the same site with unlimited generations, free MP3 downloads and commercial use included. That is a 4-billion-parameter diffusion transformer in its XL variant, not the model at the top of the leaderboard, and it is the closest thing either platform has to a no-strings option.

Access is the fork in the road

ElevenLabs exposes music through a documented API with a model identifier — music_v2_5 appears in its own pricing examples — and a request shape that takes a prompt, a length in milliseconds, and returns audio. That is the whole integration. If you are building a feature that generates a jingle on demand, or a game that scores itself, or a tool that turns a lyric sheet into a demo, ElevenLabs is available today and ACE Music v1 is not.

ACE Music v1 ships as a web application. Its documentation covers the ACE Studio desktop product rather than an inference endpoint, its product domain exposes no API routes, and no developer platform subdomain resolves. Whether that is a deliberate consumer-first strategy or a gap yet to be closed is not something the vendor has said publicly; what is certain is that as of 11 November 2026 there is no published way to call the top-ranked vocal model from a program.

This is the situation where a routing layer earns its keep, and it is worth being precise about how. OrcaRouter puts 200-plus models behind a single endpoint with provider list price passed through at 0% markup and automatic failover across providers. Music generation is not on that endpoint today, and saying otherwise would be false. What the pattern buys a team in this position is a decision that survives a vendor change: when a music model does publish a listable API, adding it is a model string next to the one already in production rather than a new contract, a new SDK and a new billing relationship. The reason this matters in a matchup with a hard access split is that the access split is the item most likely to change, and the cheapest way to prepare for it is to not have your architecture welded to either vendor.

The vocal gap in practical terms

Elo separates models by preference, and questions about preference are only useful if you know what the voters were hearing. For this pair, the difference reported by early users tracks the companies' histories. ACE Studio has spent seven years on singing voice synthesis — its Verse family of vocal models, phoneme-level editing, breath and dynamic controls, eight singing languages — and its song model inherits that. ElevenLabs came to music from speech, where its strength is intelligibility, expressiveness in short bursts, and voice identity, and its music model is strongest exactly there: a clear lead vocal that sounds like a person, over arrangements that are competent rather than inventive.

Whether that reads as a 144-point gap depends on your material. A dense pop production with a lot of vocal movement will expose the difference. A spoken-word or rap-adjacent track, a spoken intro, a hook that sits on one note — those live in the part of the problem ElevenLabs solved first, and the leaderboard margin will feel smaller than the number implies.

Choosing between them

Pick ACE Music v1 if the output is a song with a singer in it, you are working in a browser or a DAW rather than a codebase, and the price difference matters — seven dollars ninety for five hundred songs with downloads and commercial rights is close to a rounding error against any subscription music tool. The Artist tier's bundled ACE Studio access is the strongest argument in its column: a generated song you can open in a workstation, retune the vocal note by note and re-render is a workflow no other model in this comparison offers at any price.

Pick ElevenLabs Music if you need a programmatic endpoint, if your deliverables include instrumental work that has to be judged on an instrumental benchmark, or if you are already paying ElevenLabs for speech and dubbing and the marginal cost of music is effectively zero. Accept the per-track price and the self-serve licensing carve-outs as the cost of that access.

Pick neither if what you actually need is a vocal performance rather than a song. ACE Studio's own vocal synth models will do MIDI-and-lyrics singing with note-level control for a fraction of what any full-song generator costs, and that has been the company's product since 2018 — the song model is the new thing, not the thing it is best at.

The number that should stay in your head is not 1,160 or 1,016. It is that the leader on vocals cannot be called and the one that can be called is 144 Elo behind the leader. Both of those facts are true this week, and only one of them is likely to still be true in six months.