Lyria 3.5 vs Lyria 3 Pro: What's the Difference (2026)?

Two Google music models, almost the same name, and no clear signal about which one you should actually use. If you searched "Lyria 3.5 vs Lyria 3 Pro" you probably just saw a headline about Google's newest model and now you're stuck: is 3.5 a straight upgrade of 3 Pro? A different product? Do you switch, wait, or ignore it? The confusion is real because these two models come from the same family but were announced months apart, live in different places, and — crucially — have very different amounts of public documentation behind them.

Here's the honest framing this comparison is built on. Lyria 3 Pro is thoroughly documented: Google published announcements, a model card, a Vertex AI post, and a full prompting guide, so we know its length, fidelity, languages, inputs, and access paths. Lyria 3.5 is brand new (its model card is dated July 29, 2026) and much of what people are quoting about it is press coverage, not confirmed spec. So the useful question isn't "which is better on paper" — it's "what is actually confirmed about each, and which fits what you're doing today." Everything below separates documented fact from what's still rolling out, and flags the unknowns instead of guessing at them.

The Short Answer

Lyria 3 Pro is Google's production model for full, structured songs — up to three minutes with intros, verses, choruses, and bridges — available across a wide range of Google surfaces and credit-based studios right now.

Lyria 3.5 is Google DeepMind's newest music model, described in its model card as "Google's most capable music generation model to date," first surfacing in the Google Flow Music web app plus an API for downstream providers. Many of its exact specs (length, sample rate, language count, pricing) are not in the model card — treat those as still rolling out, verify before relying on them.

If you need a dependable, well-documented tool with clear access paths today, Lyria 3 Pro is the safe pick. If you want the latest fidelity and lyric handling and you're happy to work inside Flow Music while details firm up, Lyria 3.5 is the one to watch.

Lyria 3.5 vs Lyria 3 Pro: What We Actually Know

Read down the column that matches your question, not across every row. Anything marked verify is not confirmed in the official model card.

The one-line read: 3 Pro is the known quantity you can plan around; 3.5 is the newer, more capable model whose fine print is still filling in. They are not competing tiers so much as different points in a fast-moving release cycle.

If you want to run a documented model on your own brief right now, you can generate a full structured song from a single prompt box built on Lyria 3 Pro — no Google Cloud setup, no waiting for 3.5 details.

What's Confirmed About Lyria 3 Pro

Because Lyria 3 Pro has a full paper trail, you can build a workflow on it without caveats:

• Three-minute structured songs. From one text prompt, Lyria 3 Pro produces an original song up to three minutes long with intros, verses, choruses, and bridges — plus AI-generated vocals and lyrics. A single request returns the whole track, not per-30-second chunks.
• 48 kHz stereo fidelity. Per the DeepMind model card, output is high-fidelity stereo at 48 kHz — a concrete number you can cite for client or broadcast work.
• Real structure control. You can script the arrangement with timestamp cues ([00:00] ... [03:00]) so a bridge lands exactly where you want it. This is documented in Google's official prompting guide.
• Rich inputs. Text prompts, your own lyrics, up to 10 reference images, and PDF. There is no audio upload and no voice cloning — an important limit to know going in.
• Wide access. It shows up in the Gemini app, Google Vids, the Gemini API (where developers hit google/lyria-3-pro-preview), Vertex AI, ProducerAI, and credit-based studios.

For a full walkthrough of the model — prompt formula, flags, and limitations — our complete Lyria 3 Pro guide covers it in one place.

What's Confirmed (and Not) About Lyria 3.5

Stick to the model card and only a handful of things are certain:

• It is Google DeepMind's newest music model and is called its most capable to date — "a music generation system capable of synthesizing high-quality audio from a text prompt."
• It uses the same latent-diffusion architecture applied to temporal audio latents as the rest of the Lyria family.
• Inputs: text. Outputs: audio (music) plus text (lyrics). SynthID is on all outputs.
• The card claims two specific gains: significantly better audio fidelity than Lyria 2, and better prompt adherence with lyrics on both simple and complex instructions.
• Availability: Google Flow Music (a web app) and an API for downstream providers.

Now the important part — what the model card does not say, so we won't invent it. Maximum track length, exact sample rate, how many vocal languages it supports, native stem separation, and pricing are not stated for Lyria 3.5. Press coverage floats numbers (roughly three-minute songs, free access inside Flow Music, extras like conversational creation, stem separation, remixing, and Veo video) — but that is journalism, not spec. Label all of it "still rolling out — verify," and for anything unconfirmed, fall back on the documented Lyria 3 / 3 Pro behavior rather than assuming 3.5 matches it.

One thing the card notably does not confirm: whether Lyria 3.5 is in the Gemini app or Vertex AI. As of its release it points to Flow Music and downstream-provider APIs. If your plan depends on a specific Google surface, verify it's actually live there before committing.

Pick Lyria 3 Pro If…

• You need a documented, dependable model with known length, fidelity, and language support.
• You want structure control — verse/chorus/bridge with timestamp scripting — for scoring video or building a full arrangement.
• You need image-guided generation or want to feed your own lyrics in one of 8 languages.
• You want flexible access without a subscription — a credit-based studio runs full Pro generations pay-as-you-go, with no Google Cloud setup.
• You're shipping work now and can't wait for a spec sheet to settle.

Pick Lyria 3.5 If…

• You specifically want the newest, most capable model and the fidelity/lyric-adherence gains it claims over earlier versions.
• You're comfortable working inside Google Flow Music (or building on its downstream API) and don't need a guaranteed length or language count yet.
• You're experimenting rather than locking in a production pipeline, and you'll verify each spec as Google publishes it.

Rule of thumb: if you need to plan around known specs, use Lyria 3 Pro today; if you want to explore the latest model and can tolerate unconfirmed details, try Lyria 3.5 in Flow Music. For most creators shipping real deliverables this quarter, the documented model wins on predictability — you can always migrate a proven prompt to 3.5 once its fine print is public.

How to Decide Without Guessing (a Low-Friction Test)

You don't have to pick blind. Run this quick check before you commit:

1. Write one honest sentence describing the track you need: genre, mood, one lead instrument, one vocal direction (or Instrumental. for no vocals).
2. Generate it on a documented Lyria 3 Pro path first so you have a known-quality baseline — you can run that prompt through the generator in a couple of minutes.
3. If you have Flow Music access, run the same sentence through Lyria 3.5 and compare fidelity and lyric accuracy directly.
4. Ask two questions of each result: Did it nail the structure and length I need? and Are the specs I depend on actually confirmed for this model?

If 3.5 clearly sounds better and its relevant specs are confirmed for your use, switch. If either answer is shaky, stay on the documented model until 3.5's details firm up. Many creators will end up using both — 3 Pro for planned production, 3.5 for exploring the frontier.

FAQ

Is Lyria 3.5 better than Lyria 3 Pro? By its model card, Lyria 3.5 is Google's most capable music model to date, with better fidelity than Lyria 2 and better lyric adherence. But many of its specs (length, sample rate, languages, pricing) aren't published yet, while Lyria 3 Pro's are fully documented. "Better" depends on whether the specs you rely on are confirmed for 3.5 — verify before you switch.

Is Lyria 3.5 a replacement for Lyria 3 Pro? Not officially. The 3.5 model card points to Google Flow Music and a downstream-provider API, and does not confirm Lyria 3.5 in the Gemini app or Vertex AI where 3 Pro lives. Treat them as coexisting for now, not a one-for-one upgrade.

Do Lyria 3.5 and Lyria 3 Pro use the same architecture? Yes — both use latent diffusion applied to temporal audio latents, the shared Lyria family design. The differences are in capability and where each model is available, not the underlying approach.

How long are songs in Lyria 3.5 vs 3 Pro? Lyria 3 Pro is documented at up to 3 minutes per generation. Lyria 3.5's maximum length is not stated in its model card; press reports suggest around three minutes but that's unconfirmed — verify against Google's current documentation.

Where can I use each one? Lyria 3 Pro is available in the Gemini app, Google Vids, the Gemini API, Vertex AI, ProducerAI, and credit-based studios. Lyria 3.5 (per its model card) is in Google Flow Music plus an API for downstream providers. Availability on other surfaces isn't confirmed for 3.5 yet.

Which should I use today? If you're shipping real work, start with the documented model — Lyria 3 Pro — because you can plan around its known specs. Experiment with Lyria 3.5 in Flow Music alongside it, and migrate once its details are official.

Bottom Line

Lyria 3.5 and Lyria 3 Pro aren't rival tiers — they're two moments in one fast-moving release cycle. Lyria 3 Pro is the model you can build on today: three-minute structured songs, 48 kHz stereo, 8 vocal languages, image inputs, and access almost everywhere, all confirmed on paper. Lyria 3.5 is the newer, more capable model whose headline claims (fidelity and lyric adherence) are official but whose length, languages, and pricing are still rolling out — so verify before you depend on any of them.

The practical move is to anchor on the documented model now and keep an eye on 3.5 as Google fills in the blanks. Write one sentence describing the song you need and create it with the Lyria 3 Pro generator — you'll get a known-quality track today, and a baseline to judge Lyria 3.5 against the moment its specs are confirmed.

Sources

Primary sources for the claims above. Lyria 3.5 details not present in its model card are labeled "verify" in the text — confirm them against Google's current documentation, as both models are partly in preview and specs can change after publication (July 2026).

• Google DeepMind: Lyria 3.5 model card — newest model, architecture, fidelity and lyric-adherence gains, SynthID, availability
• Google DeepMind: Lyria 3 model card — architecture, 48 kHz output, SynthID
• Google: Lyria 3 Pro announcement — release date, 3-minute structured songs, availability
• Google DeepMind: Lyria model page — model family and capabilities
• Google Cloud: Lyria 3 and Lyria 3 Pro on Vertex AI — inputs (images/PDF), preview status, C2PA
• Google Cloud: Ultimate prompting guide for Lyria 3 Pro — prompt formula, timestamp technique, languages
• TechCrunch: Google launches Lyria 3 Pro — launch coverage