Lyria 3.5 vs Suno v4: Two Years Apart, One Category — and the Bar Has Already Moved

DeepMind launched Lyria 3.5 on July 29, 2026 inside Flow Music, and the comparison everyone reaches for is Suno v4 — the November 2024 model that turned Suno into a category. That instinct is a time-lapse photograph. Suno v4 is 21 months old in a market that has since shipped v4.5, v5, and v5.5; Lyria 3.5 is ten days old, reachable only inside a DeepMind product, and carries no benchmark score of its own. On the one independent scoreboard that tracks this space — the Artificial Analysis Music Arena, where real listeners vote in blind tests — neither model in this headline even appears: Suno v4 is too old to be listed, and Lyria 3.5 is too new to be callable.

That absence is not a dead end; it is the point. The nearest measured stand-ins — Lyria 3 Pro, the model Lyria 3.5 replaces, and Suno v4.5, the model that closed v4's gaps — are statistically tied on vocals at 1,088 to 1,083 Elo, while Lyria leads clearly on instrumentals at 1,118 to 1,083. So the honest shape of this matchup is: against the generation that made Suno famous, Google has effectively caught up, and on some dimensions passed it; against the Suno of today, the bar has already moved past both of them.

The models, two years apart

Suno v4 shipped November 19, 2024 as the model that completed Suno's pitch: type a description of genre, mood, and theme, and get back a full four-minute song with coherent verse-chorus-bridge structure. It brought the ReMi lyric writer, the Personas feature that carries a vocalist's sound across tracks, Remaster for upgrading older generations, and Cover for reinterpreting existing audio. Its vocals were the reference point for the whole industry — reviewers called them the cleanest, most human AI vocals of the time — and it made Suno the consumer leader, a position it has held since.

Lyria 3.5 is Google's counter-move, 21 months later. It is the third Lyria in five months — Lyria 3 shipped in February with 30-second clips in the Gemini app, Lyria 3 Pro followed in March with three-minute songs, and 3.5 arrived July 29 inside Flow Music. Google's upgrade list is its own: richer, more natural melodic structures; lyrics with better prompt adherence; more emotionally nuanced vocals with improved pronunciation; and direct control over tempo and duration. What Google did not publish is any number — no listening-test score, no efficiency figure, no head-to-head. Quality is asserted, not measured, which matters when the model you are comparing it to built its reputation the other way, on millions of real generations.

The dimension-by-dimension contrast

• Released — Lyria 3.5: July 29, 2026 / Suno v4: November 2024.

• Where it lives — Lyria 3.5: Google Flow Music, free tier plus $6/$18/$48 plans / Suno v4: Suno's app, Free with 50 credits a day, Pro $10, Premier $30 a month.

• Track length — Lyria 3.5: 30 seconds to 3 minutes / Suno v4: up to about 4 minutes.

• Vocals — Lyria 3.5: eight languages, SynthID-watermarked, more expressive delivery (vendor claim, unverified) / Suno v4: the vocals that made Suno famous, still the era's reference.

• Creative control — Lyria 3.5: Selective Section Painting, tempo/BPM/key/duration, exportable vocal, drum, and bass stems / Suno v4: Extend, Cover, Personas, Remaster; no native stem export.

• Independent score — Lyria 3.5: none (proxy Lyria 3 Pro: vocals 1,083, instrumental 1,118) / Suno v4: none (proxy Suno v4.5: vocals 1,088, instrumental 1,083), Artificial Analysis Music Arena Elo.

The honest scoreboard: neither of these two is on it

Artificial Analysis runs the only public, independent, blind-preference arena for music generation — the Music Arena — and it is where this matchup's real numbers live. On vocals, Suno V5.5 leads at 1,171 Elo, Suno V5 sits at 1,097, Suno V4.5 at 1,088, and Lyria 3 Pro at 1,083. On instrumentals, Suno V5.5 leads at 1,188, Lyria 3 Pro comes in at 1,118, and Suno V4.5 at 1,083.

Two models you might expect are missing. Suno v4 does not appear — it predates the arena and was superseded before it mattered. Lyria 3.5 does not appear either, because it is not reachable by any API; you cannot run it through a harness, so no one has scored it. That is why the proxies carry the argument. Lyria 3 Pro is Lyria 3.5's direct predecessor, and Google's claims for 3.5 are all improvements on it. Suno v4.5 is v4's direct successor, the version that fixed v4's weaknesses. Set those two against each other and you get the honest 2026 picture: vocals statistically tied, instrumentals clearly Google's.

Price, and the API question neither company answers

The pricing math is the part developers should actually copy down. Flow Music is free on signup with 500 credits, then Starter at $6, Plus at $18, and Member at $48 a month, with Google AI Plus, Pro, and Ultra subscribers folding in the matching tier. Suno runs Free at 50 credits a day, Pro at $10 a month for roughly 500 songs, and Premier at $30 a month for roughly 2,000 — which on the top plan works out to about a cent and a half per finished song, subscription economics that are hard to beat for volume.

On the per-track API side, Google prices the Lyria family at roughly $0.04 for a 30-second Lyria 3 clip and about $0.08 for a full three-minute Lyria 3 Pro song — vendor-reported figures for the previous generation, not yet confirmed for 3.5. Suno v4 has no official API at all; developers who want programmatic Suno generation get v4- and v4.5-era models through third-party wrappers at anywhere from roughly $0.06 to $0.55 per generation, with no official price list, no SLA, and no guarantee the endpoint stays up. And Lyria 3.5 has no confirmed developer surface either — Google has not said whether it replaces Lyria 3 Pro on the Gemini API or AI Studio.

So both companies are selling consumer products that developers want, and neither has a first-party endpoint to point at. That is the class of problem a routing layer exists for. When a music model does land behind a real API — a Lyria 3.5 endpoint, or the official Suno API that Suno said in July it is exploring — OrcaRouter passes the provider's list price straight through at 0% markup, so a vendor price cut is live on our side the same day, and automatic failover is the safe way to put a brand-new, unproven music model on a production path beside a proven one. We don't route Lyria 3.5 or Suno v4 today and we won't pretend otherwise — neither is on a first-party API we can call. But the per-track math above is exactly the number a routing layer keeps honest.

Creative control: the thing v4 never had

The sharpest product difference between a 2024 model and a 2026 model isn't audio quality — it's editing. Suno v4's tools were structural: Extend to lengthen a clip, Cover to re-imagine an existing song, Personas to carry a vocal style forward, Remaster to lift older tracks. What you could not do was reach into a finished song and fix one part of it. A bad chorus meant regenerating the whole track, and regenerations are probabilistic, so you might not get the chorus back at all.

Lyria 3.5's headline feature is exactly that: Selective Section Painting, which reworks a specific part of an existing track — the bridge, say — or extends a short melody into a full song without starting over. It pairs that with direct tempo, BPM, key, and instrumentation controls, vocals in eight languages, and separate vocal, drum, and bass stems built for clean stem-splitting workflows. The DeepMind model page lays all of this out, including the SynthID watermark that rides on every track.

The legal backdrop is part of the choice

Legal posture belongs in this comparison. Suno has been in litigation with the labels since the v4 era, and the fights are still running: Sony Music's case against Suno is active, and Suno lost a first-instance ruling in Munich on July 31, 2026. Google, meanwhile, claims it trains on data it and YouTube have rights to, but independent musicians sued over Lyria 3 in March 2026 — the complaint alleges "at least 44 million clips and 280,000 hours of music" pulled from YouTube — and Google's motion to dismiss, filed in June, is still pending. Lyria 3.5 tracks carry SynthID, giving verifiable provenance, though Google mentions it only on the model page, not in the launch post. Neither company is a clean "licensed" choice right now; both have open questions.

Who should pick which

Choose Lyria 3.5 if you want the better productized experience today: free to start in Flow Music, section-level editing that Suno v4 never had, tempo and key control, eight-language vocals, and a SynthID watermark on everything you export. By proxy it clears the v4-generation bar — its predecessor already ties Suno v4.5 on vocals and beats it on instrumentals, and Google claims 3.5 goes further. The constraints are real: three minutes max, one-shot generation, and no confirmed API.

Choose Suno v4 if you specifically want the proven vocal reference, longer single-pass songs up to about four minutes, and the largest community and shared-track ecosystem in the category. But be honest about what you are choosing: v4 is a 2024 model. Suno has shipped v4.5, v5, and v5.5 since, and the current leaderboard puts v5.5's vocals at 1,171 Elo — about 80 points above v4.5's. If the Suno that interests you is the one with voice cloning and a built-in DAW, that is v5.5, not v4.

The verdict is a time-lapse too. If this were actually the 2026 fight, you would line Lyria 3.5 up against Suno v5.5, not v4. The reason the v4 comparison still earns its place is that it shows how far Google has come in under two years — from a model that could not finish a song to one that, on the evidence that exists, matches the generation that made Suno a category. It just arrived after the category had already moved.

FAQ

Why compare a 2026 model to a 2024 one?

Because Suno v4 is the version people still picture when they think "Suno," the one several third-party music APIs still serve, and the quality bar Google's claims are implicitly measured against. But it is not Suno's current model — v4.5, v5, and v5.5 have shipped since, and the actual 2026 comparison is Lyria 3.5 vs Suno v5.5. This piece treats v4 as the baseline generation, which is what it is.

Which one actually sounds better?

There is no direct test: Suno v4 is off the leaderboard and Lyria 3.5 has no API, so no one has run them head-to-head. The closest evidence is Artificial Analysis' Music Arena, where Lyria 3 Pro and Suno v4.5 are statistically tied on vocals (1,083 vs 1,088 Elo) and Lyria leads on instrumentals (1,118 vs 1,083). Google claims 3.5 improves vocals further. Listen with your own ears; the numbers say the gap Google once had to close is essentially closed.

Can I call either of these from code?

Not officially. Suno has no developer API — programmatic access runs through third-party wrappers at roughly $0.06 to $0.55 per generation. Lyria 3.5 is only in Flow Music; Google has not confirmed an API, pricing, or whether it replaces Lyria 3 Pro on the developer platform. Neither model is on OrcaRouter, because neither has a first-party endpoint to route to.