You just saw "Lyria 3.5" flash across your feed — a new Google music model, dropped on July 29, 2026 — and now you have questions the headlines refuse to answer. Is it a whole new product or a point update? Did it replace the Lyria 3 Pro you may already be using? Where do you actually go to try it, and can you get it inside the same apps as before? Most of the coverage is either breathless hype or a copy-paste of Google's press line, and neither tells you what genuinely changed or what is still vapor.
Here is the trap this post exists to fix: with a name like "3.5," everyone assumes it is a small, drop-in upgrade to Lyria 3 that instantly shows up everywhere. It is not that simple. Lyria 3.5 is a distinct, newer model with real improvements Google has documented — but a lot of the specs people are quoting online are not in the official model card, and pretending otherwise will only set you up for disappointment. So this guide sticks strictly to what Google DeepMind has actually confirmed, flags everything that is still rolling out, and shows you where each model fits.
What Lyria 3.5 Actually Is
Lyria 3.5 is Google DeepMind's newest AI music generation model, published on its official model card dated July 29, 2026. In Google's own words, it is a "music generation system capable of synthesizing high-quality audio from a text prompt," and the card calls it "Google's most capable music generation model to date."
Strip away the marketing and it does what the Lyria family has always done: you describe the music you want in plain text, and the model composes and renders audio. Under the hood it uses the same architecture as the rest of the family — latent diffusion applied to temporal audio latents — so this is an evolution within a known lineage, not a ground-up reinvention. Its inputs are text; its outputs are audio (music) plus text (lyrics). And like every Lyria output, everything it generates carries a SynthID invisible watermark marking it as AI-generated.
The short version: same family, newer and more capable model, text in, music-and-lyrics out.
What Changed vs Lyria 3 (Confirmed on the Model Card)
This is the part worth getting exactly right, because it is where speculation runs wild. Google's model card is explicit about two headline improvements — and notably benchmarks them against Lyria 2, the previous generation, rather than Lyria 3:
- Audio fidelity. Lyria 3.5 "improved significantly compared to Lyria 2 on audio fidelity." In plain terms: cleaner, higher-quality sound.
- Lyric prompt adherence. It follows lyric instructions — "simple and more complex" alike — more accurately. If you feed it structured or detailed lyric prompts, it is meant to stick to them more faithfully.
That is the confirmed list. Everything else you have read about Lyria 3.5 — bigger jumps in musicality, more expressive vocals, finer creative control over tempo and duration — comes from press coverage, not the model card, so treat those as "verify" claims rather than settled facts. They may well be true; they are just not something Google has documented, and this post will not pretend they are.
What's Confirmed vs Still Rolling Out
The single most useful thing you can do with Lyria 3.5 right now is separate what Google has committed to from what is still unclear. Here is the honest split.
Confirmed by the model card:
- It exists, dated July 29, 2026, and is described as Google's most capable music model to date.
- Architecture: latent diffusion on temporal audio latents.
- Inputs: text. Outputs: audio (music) and text (lyrics).
- Improvements: significantly better audio fidelity vs Lyria 2; better lyric prompt adherence.
- SynthID watermarking on all outputs.
- Availability: Google Flow Music (a web app) and an API for downstream providers.
Not stated in the 3.5 model card — do not assume it:
- Maximum song length. The card gives no duration cap. Press reports suggest around three minutes, but that is unconfirmed — verify.
- Vocal languages. No language list. Lyria 3 Pro documents eight languages; whether 3.5 matches that is not yet detailed.
- Sample rate. No figure. Lyria 3 Pro documents 48 kHz stereo; 3.5's spec is not published.
- Native stem separation. Not mentioned in the card. Some coverage of Flow Music mentions stem tools — treat as "verify."
- Pricing. No price. Coverage says it is free to Flow Music users, but there is no unified public rate card. Verify before you budget around it.
- Gemini app / Vertex AI support. The card names only Flow Music and the downstream API. It does not confirm the Gemini app or Vertex AI for 3.5, so do not assume 3.5 is live on the paths you used for Lyria 3 Pro.
The practical takeaway: for anything the 3.5 card does not spell out, the safest move is to fall back on the well-documented Lyria 3 / 3 Pro facts and assume the 3.5 details are still being finalized.
Lyria 3.5 vs Lyria 3 Pro — What We Know
Because the names are so close, this is the comparison everyone actually wants. This table sticks to what is documented for each model and marks the gaps honestly. (For a deeper side-by-side once more 3.5 specs land, see our dedicated Lyria 3.5 vs Lyria 3 Pro breakdown.)
| Lyria 3.5 | Lyria 3 Pro | |
|---|---|---|
| Positioning | Newest; "most capable to date" | Full-song production model |
| Released | July 29, 2026 (model card) | Announced March 25, 2026 |
| Architecture | Latent diffusion (same family) | Latent diffusion (same family) |
| Song length | Not stated (press ~3 min, verify) | Up to 3 minutes, documented |
| Vocal languages | Not stated | 8 languages, documented |
| Sample rate | Not stated | 48 kHz stereo, documented |
| Confirmed availability | Flow Music + API for providers | Gemini app, Google Vids, AI Studio/API, Vertex AI, credit studios |
| Watermark | SynthID | SynthID |
Read that table the right way: Lyria 3 Pro is the model with a fully documented, broadly available feature set you can rely on today. Lyria 3.5 is newer and, by Google's own description, more capable — but much of its spec sheet is still blank. "Newer" and "more usable for your exact job right now" are not the same thing.
Where to Actually Use Lyria 3.5 — and What to Do Today
Per the model card, Lyria 3.5 lives in two places: Google Flow Music, a web app, and an API for downstream providers. That is a narrower footprint than Lyria 3 Pro, which is reachable through six different paths. So if you were hoping to fire up 3.5 inside the Gemini app you already pay for, pause — that is not confirmed yet.
This is the awkward gap between "a new model was announced" and "I can use it for my project this afternoon." Announcements are instant; broad, documented availability takes time to roll out. If you need finished, structured songs today — with a known length, known languages, and a predictable cost — the pragmatic answer is to work with the model that is already fully documented and widely accessible.
That is exactly what a pay-as-you-go credit studio is for. You can generate full Lyria 3 Pro songs right now, with no Google subscription and no Cloud setup, using our AI music generator — up to three-minute tracks with real structure, vocals, and 48 kHz-class output. When Lyria 3.5's specs and access paths firm up, you will already know your prompt workflow. There is no reason to sit on your hands waiting for a spec sheet to fill in.
If you are brand new to the family, our explainer on what Lyria 3 is and how to reach it is the fastest way to get oriented, and you can check exact per-generation costs on the pricing page instead of guessing at a monthly tier.
Rule of thumb: treat Lyria 3.5 as the exciting "watch this" model and Lyria 3 Pro as the "ship it today" model. Follow 3.5's rollout, but produce on what is documented and available.
How to Tell Whether You're Really On Lyria 3.5
Because the naming is confusing and access is uneven, it is easy to think you are using 3.5 when you are not. A quick, low-friction check: confirm the surface first. If you are inside Google Flow Music or hitting an API endpoint that a provider explicitly labels as Lyria 3.5, you are on the new model. If you are in the Gemini app, Google Vids, or a general credit studio, you are almost certainly on Lyria 3 or Lyria 3 Pro — which is completely fine, and in most cases the more documented, more predictable choice today. When in doubt, ask the platform which model version it routes to rather than assuming the newest name.
Frequently Asked Questions
What is Lyria 3.5 in one sentence? Lyria 3.5 is Google DeepMind's newest AI music model — a text-to-music system its model card calls "Google's most capable music generation model to date," available via Google Flow Music and an API for downstream providers.
When was Lyria 3.5 released? Its official model card is dated July 29, 2026. That is the date to trust; other timelines floating around are unconfirmed.
How is Lyria 3.5 different from Lyria 3? The model card confirms two improvements — significantly better audio fidelity than Lyria 2, and more accurate lyric prompt adherence. Other differences you may read about (musicality, vocal expressiveness, creative control) come from press coverage, not Google, so verify them.
How long can Lyria 3.5 songs be, and what languages does it support? Not yet detailed. The 3.5 model card does not state a maximum length, a language list, or a sample rate. Lyria 3 Pro documents up to three minutes, eight vocal languages, and 48 kHz stereo — but you should not assume 3.5 matches those numbers until Google publishes them.
Is Lyria 3.5 free? There is no unified public price. Some coverage says it is free to Google Flow Music users, but that is unverified and Google publishes no single rate card. If you want a known cost today, a pay-as-you-go credit studio running Lyria 3 Pro gives you transparent per-generation pricing.
Can I use Lyria 3.5 in the Gemini app or on Vertex AI? Not confirmed. The 3.5 model card names only Google Flow Music and the downstream-provider API. It does not list the Gemini app or Vertex AI, so treat those as unavailable for 3.5 until Google says otherwise.
The Bottom Line
Lyria 3.5 is a real, newer, genuinely more capable model — Google DeepMind says so on the model card — with documented gains in audio fidelity and lyric adherence, delivered through Google Flow Music and a provider API, and watermarked with SynthID like everything else in the family. What it is not, yet, is a fully specced product: no confirmed length, language count, sample rate, stem support, or price, and no confirmed Gemini or Vertex AI availability. Anyone quoting those numbers is quoting the press, not Google.
So follow the 3.5 rollout with interest — but if you need finished songs today, produce on the model that is already documented and everywhere. Generate full Lyria 3 Pro tracks on demand with our credit-based studio, no subscription required, and check the pricing page for exact costs while you wait for 3.5's spec sheet to fill in.
