Google just quietly shipped something that the music production world has been watching for a while. Lyria 3.5, the company’s most capable AI music generation model to date, is now available inside the Gemini app and through the Gemini API — and if you’ve followed where this technology was six months ago, the jump in quality is genuinely striking. More expressive vocals. Richer arrangements. Better overall fidelity. This isn’t a minor patch release.
How We Got Here: The Road to Lyria 3.5
Google’s been in the AI music space longer than most people realize. The original Lyria model debuted in late 2023, initially powering YouTube’s Dream Track experiment — a feature that let creators generate short musical clips in the style of certain artists. It was impressive for the time, but it sounded like what it was: an early demo. Thin instrumentation, vocals that felt slightly off, arrangements that lacked dynamic range.
Lyria 2 arrived in 2024 with meaningful improvements to instrumental generation, and Google opened it up to more developers through the API. The model started showing up in third-party tools, Google’s own AI Test Kitchen, and some early integrations with YouTube creator features. But vocals remained a weak point — the kind of thing you’d flag if you played a clip to any working musician.
Lyria 3.5 is the version that feels like it’s meant to compete seriously. Google’s announcement on September 4, 2026 positions it as the best-sounding model they’ve built — and based on what’s been shared publicly, the vocal and arrangement upgrades are the headline story.
What’s Actually New in Lyria 3.5
Let’s get specific, because “more expressive” is the kind of phrase that means nothing without context.
Vocal Quality Takes a Real Step Forward
The most talked-about improvement is in vocal generation. Earlier Lyria versions struggled with what musicians call phrasing — the way a singer naturally breathes, emphasizes syllables, and bends into notes. It’s subtle, but it’s also exactly what separates a track that sounds human from one that sounds synthetic.
Lyria 3.5 reportedly handles this significantly better. The model appears to have been trained on a much larger and more diverse vocal dataset, and the results reflect that. Emotional nuance — the slight rasp on a low note, the breath before a chorus — comes through in a way that earlier versions couldn’t manage.
Richer Musical Arrangements
The second big upgrade is in how the model handles full arrangements. Previous versions of Lyria tended to flatten dynamics — everything sat at roughly the same energy level, which made longer tracks feel monotonous. Lyria 3.5 introduces what Google describes as more sophisticated musical structure awareness. Intros build. Choruses hit harder. Bridges actually feel like bridges.
For anyone who’s tried to use AI music tools for anything beyond a 30-second clip, this matters a lot. Suno and Udio — the two biggest competitors in this space — have been working on the same problem. Whether Lyria 3.5 pulls ahead of them on arrangement quality is something that’ll take real-world testing to determine, but Google is clearly taking aim at that benchmark.
Key Features at a Glance
- Expressive vocal generation with improved phrasing, breath, and emotional range
- Dynamic musical arrangements with better structural awareness across intros, verses, choruses, and outros
- Available in the Gemini app directly — no separate tool required
- Gemini API access for developers building music-integrated applications
- Multi-genre support spanning pop, R&B, electronic, folk, and more
- Text-to-music prompting with natural language inputs
- Higher audio fidelity compared to Lyria 2 outputs
Pricing and Availability
As of the September 4 release, Lyria 3.5 is live in the Gemini app for users on Gemini Advanced (the paid tier, currently $19.99/month as part of Google One AI Premium). API pricing follows Google’s standard Gemini API structure — developers working at scale will want to check the Gemini API pricing page directly, as music generation endpoints are typically metered differently than text or image calls.
How Lyria 3.5 Compares to the Competition
The AI music generation space has gotten crowded fast. Here’s the honest picture.
Suno and Udio Are the Benchmarks to Beat
Suno v4 has arguably set the consumer standard for AI music. It’s fast, it’s accessible, and it produces tracks that regularly fool casual listeners. Udio carved out a slightly different niche — more control for musicians who know what they want, better at genre-blending, and with a vocal model that’s been competitive with anything else on the market.
Both companies have raised significant funding and built meaningful user bases. Suno in particular has become the default recommendation when someone asks “what’s the best AI music tool right now?” That’s the position Google is trying to chip away at with Lyria 3.5.
Google’s edge isn’t just the model quality — it’s distribution. Lyria 3.5 sitting inside Gemini means it’s immediately accessible to millions of Gemini users who were never going to sign up for a standalone music app. That’s a different kind of competition than going head-to-head on features alone.
What About OpenAI?
OpenAI doesn’t have a public music generation product. They’ve been rumored to be working in this direction, but nothing’s shipped. That leaves a genuine opening for Google to establish Lyria as the default AI music layer — especially given the API availability, which means developers can build on it right now. For more on where OpenAI’s AI capabilities currently stand, our breakdown of GPT-6 Astra’s capabilities gives useful context on where they’re focusing their energy.
What This Means for Different Audiences
For Casual Creators and Content Producers
If you’re making YouTube videos, podcasts, social content, or anything that needs background music, Lyria 3.5 is a practical upgrade. The ability to generate tracks directly in Gemini — using the same interface you’re already using for other tasks — removes friction. You describe what you want, you get a track, you move on. No separate subscription to manage.
The improved vocal quality also opens up options that weren’t really viable before. Generating a full song with lyrics and vocals for a short-form video? That’s now a realistic workflow for non-musicians.
h3>For Developers and Builders
The API access is the real story for the developer community. Lyria 3.5 through the Gemini API means you can embed music generation into apps, games, interactive experiences, or production tools without building the model infrastructure yourself. Think about what that enables: a video editing app that auto-generates a score based on the mood of your footage. A game that creates adaptive background music based on player state. A meditation app that generates personalized ambient tracks on demand.
The precedent for this kind of AI-powered workflow acceleration is already well-established — we’ve seen it happen with text, image, and now video generation. Music was the last major creative domain without a serious API-accessible solution from a major platform. That gap just got smaller. This mirrors what we’ve seen with Google’s broader push to ship multi-agent AI capabilities that developers can actually build on top of.
For Working Musicians
This is the complicated one. AI music tools have been a source of genuine anxiety in the music industry — and that anxiety is legitimate. Lyria 3.5 doesn’t resolve those questions. It intensifies them. The model producing expressive vocals with good arrangement quality is exactly what people worried about when they worried about AI music.
At the same time, some working musicians have found real utility in these tools for ideation — sketching out a chord progression, testing a melody idea, generating a reference track to communicate a direction to a collaborator. Whether Lyria 3.5 becomes a tool for musicians or a replacement for them probably depends more on how the industry responds than on the technology itself.
Key Takeaways
- Lyria 3.5 is Google’s most capable music generation model yet, with meaningful upgrades to vocal expressiveness and arrangement quality
- It’s available now in the Gemini app for Advanced subscribers and through the Gemini API for developers
- The main competitors are Suno and Udio — Google’s distribution advantage through Gemini is a significant factor independent of model quality
- API access opens up a new category of music-integrated applications that weren’t previously practical to build
- For casual creators, it’s a low-friction way to generate original music without a standalone subscription
- The broader questions around AI music and the creative industry aren’t answered here — they’re amplified
Frequently Asked Questions
What is Lyria 3.5 and who made it?
Lyria 3.5 is Google’s latest AI music generation model, developed by Google DeepMind. It’s designed to generate full musical tracks — including vocals and instrumentation — from natural language text prompts, and is now integrated directly into the Gemini app and Gemini API.
How does Lyria 3.5 compare to Suno and Udio?
All three are capable AI music tools, but they serve slightly different contexts. Suno has the largest consumer user base and is optimized for speed and accessibility. Udio offers more control for musicians. Lyria 3.5’s advantage is Google’s distribution — it lives inside Gemini, which means immediate access for a massive existing user base without needing a separate app or account.
Is Lyria 3.5 available for free?
Access within the Gemini app requires a Gemini Advanced subscription, which is $19.99/month as part of Google One AI Premium. API access is available to developers through the Gemini API, with pricing based on usage volume — check Google’s official API pricing page for current rates.
Can developers build products using Lyria 3.5?
Yes — that’s one of the bigger stories here. The Gemini API now includes Lyria 3.5 access, meaning developers can integrate AI music generation into apps, games, and creative tools. Google AI for Developers is the starting point for anyone who wants to build on top of it.
Google has now shipped a credible, API-accessible music generation model that lives inside one of the most widely used AI platforms on the planet. Whether Lyria 3.5 becomes the industry standard or a stepping stone to something better is still an open question — but the pace of improvement here has been faster than most people expected. I wouldn’t be surprised if we see Lyria integrated into YouTube’s creator tools in a much more prominent way within the next six months. The pieces are clearly in place.
