Lyria 3.5
Google DeepMind's July 2026 music model, with better musicality, vocals and lyric handling, shipped first inside Google Flow Music.
Updated 2026-07-30
Overview
Lyria 3.5 is Google DeepMind's latest text-to-music model, announced on July 29, 2026 and rolled out immediately inside Google Flow Music. It generates full musical arrangements — instrumentation, structure, vocals and lyrics — from text prompts, and Google positions the release around three specific upgrades over earlier Lyria versions: musicality (arrangements that hold together over a full track rather than looping a good eight seconds), vocal and lyric quality, and finer creative controls over what comes out.
The important thing to understand is what Lyria 3.5 is. It's not an app you sign up for. It's the engine underneath Google's music surfaces — Flow Music first, with the broader Lyria family historically reaching developers through the Gemini API and Vertex AI and creators through Music AI Sandbox and MusicFX. That makes the comparison to Suno and Udio slightly lopsided: those are consumer products with libraries, sharing and a subscription; Lyria is a model that Google wraps in whichever product you happen to be using. If Flow is already part of your workflow, Lyria 3.5 arrives for free. If it isn't, there's no standalone "Lyria" to buy.
Like every Lyria release, output carries SynthID, DeepMind's imperceptible audio watermark, so generated tracks stay machine-detectable downstream. That's a real differentiator for anyone who needs provenance — agencies, platforms, publishers — and a mild annoyance for anyone hoping to pass AI music off as session work. Google has not published a full spec sheet or independent benchmark for 3.5 at launch, so claims about generation length, sample rate and how it stacks up against Suno's latest are, for now, unverified.
Key features
Vocals and lyrics
Google calls out lyric handling and vocal performance as headline improvements in 3.5 — historically the weakest part of model-generated music, where phrasing drifts and consonants smear. Worth judging by ear before committing to a project.
Creative controls
The release emphasizes finer control over the generation rather than pure prompt-and-pray, which matters if you're trying to hit a specific brief instead of collecting happy accidents.
Flow Music integration
Shipped into Google Flow Music on day one, so it sits next to Google's video generation tooling rather than in a separate app — convenient if you're scoring AI-generated video in the same place.
SynthID watermarking
Every Lyria output is watermarked with DeepMind's SynthID audio marker, which survives common transformations and keeps generated music identifiable as generated. Non-optional, by design.
Pricing
Free tier: Yes — Google Flow Music includes free access to Lyria generation, with usage limits set by Google's AI plan tiers rather than a separate music quota.
| Plan | Price | What's included |
|---|---|---|
| Free (via Google Flow Music) | $0 | Limited generation credits inside Flow Music; SynthID watermarking applied to all output |
| Google AI paid tiers | Check website for current pricing | Higher generation limits and priority access; Lyria is bundled into Google's AI subscription plans rather than sold separately |
| Developer access (Gemini API / Vertex AI) | Check website for current pricing | The Lyria family has been offered to developers through Google's API surfaces on usage-based pricing; confirm 3.5 availability and rates on Google's docs |
Limited generation credits inside Flow Music; SynthID watermarking applied to all output
Higher generation limits and priority access; Lyria is bundled into Google's AI subscription plans rather than sold separately
The Lyria family has been offered to developers through Google's API surfaces on usage-based pricing; confirm 3.5 availability and rates on Google's docs
Pros & cons
Pros
- ✓Free to try inside Google Flow Music — no separate subscription to evaluate the model
- ✓SynthID watermarking gives verifiable provenance, which matters for commercial and platform use
- ✓Sits alongside Google's video tooling, so scoring generated footage happens in one place
- ✓Backed by DeepMind's research pipeline, with a track record of steady Lyria iteration rather than one-off releases
Cons
- ×Not a standalone product — you get it through Flow Music or an API, not a dedicated music app with a library and sharing
- ×No published spec sheet or independent benchmark at launch, so quality claims over Lyria 3 are Google's word for now
- ×Suno and Udio have far more mature consumer workflows for song iteration, stems and community sharing
- ×Commercial-usage rights depend on which Google surface and plan you're on, which is less clear-cut than dedicated music tools' licensing
How it compares
| Tool | Best for | Pricing | Score |
|---|---|---|---|
| Lyria 3.5 | — | Free tier via Google Flow Music; paid access through Google AI subscriptions — check website for current pricing | 8.4/10 |
| Suno AI vs Suno AI → | Content creators, podcasters, and indie artists who need original full songs with vocals and lyrics without studio production. | Freemium | 9.2/10 |
| ElevenLabs vs ElevenLabs → | Creators, publishers, and developers who need realistic voice cloning or text-to-speech across 32+ languages for narration, dubbing, or apps. | Free tier + Starter $5/mo + Creator $22/mo + Pro $99/mo + Scale $330/mo + Enterprise custom | 9.2/10 |
| Udio | Music producers and enthusiasts who want radio-ready tracks with precise genre reproduction and detailed control over lyrics and structure. | Freemium | 8.8/10 |
Compare head-to-head
Comparison explorer Put Lyria 3.5 up against any three tools Opens with Lyria 3.5 already loaded. Add up to three more from the full index and read pricing, features, pros and cons in one table. →Related reading
Two Settings Tripled GPT-5.6 Sol's ARC-AGI-3 Score
OpenAI says two Responses API settings took GPT-5.6 Sol from 13.3% to 38.3% on ARC-AGI-3's public set. Where that number holds, and where it doesn't.
OpenAI Codex Security CLI: Setup and the Catch
OpenAI open-sourced the Codex Security CLI on July 29. What it scans, what to check before wiring it into CI, and who is grading whose homework.
Pacing the Frontier: 1,100 Lab Staff Sign On
A petition at pacingthefrontier.com drew 1,100+ signatures from OpenAI, Anthropic, Google and Meta staff. What it asks for, and what the count hides.
Ready to try Lyria 3.5?
Head to the official site to start with Lyria 3.5 — pricing and plans are listed above.
Visit Lyria 3.5