Google Lyria 3 Pro Is Now Available on Vertex AI — Here's What Developers Get
Google has opened Lyria 3 Pro and Lyria 3 Clip to enterprise developers via Vertex AI, marking the first public API access to the Lyria 3 model family.
Update, 23 June 2026: Gemini 3.5 Pro enters its confirmed GA window — still no public launch announcement as of June 23
Gemini 3.5 Pro is inside its publicly expected GA window (June 23–30) but has not launched. As of June 23, the official Gemini API changelog contains no GA entry for the model. Access remains limited to a small Vertex AI enterprise preview; the model is not available in Google AI Studio’s public model picker, the consumer Gemini app, or the general API as a stable release.
This matters for the Lyria 3 story because both launches sit under the same Vertex AI enterprise push Google has been running since mid-June. Lyria 3 Pro completed its rollout cleanly; Gemini 3.5 Pro has not.
Confirmed pre-GA specs include a 2-million-token context window and a Deep Think reasoning mode. Pricing remains unannounced and will only be confirmed when the model card drops at GA. The ~$15/$60 per million tokens figure circulating in third-party coverage is an estimate based on prior Pro-to-Flash pricing ratios, not a Google-confirmed number.
For developers: do not hardcode gemini-3.5-pro in production code yet. Google typically ships an initial release under a preview-suffixed identifier before stabilizing the clean model string. Watch the Google AI Studio model picker and the official changelog for the GA drop. Gemini 3.5 Flash remains the live production option in the meantime.
Google has made two Lyria 3 models available in public preview on Vertex AI: lyria-3-pro-preview and lyria-3-clip-preview. This is the first time the Lyria 3 model family has been accessible outside a closed research context, and it opens enterprise-grade AI music generation to developers at API scale.
If you’ve been watching the Lyria story from the sidelines, this is the moment it becomes practically relevant for building products.
Two Models, Two Use Cases
The two models are meaningfully different in what they’re designed to do.
Lyria 3 Clip (lyria-3-clip-preview) generates up to 30 seconds of audio at 48kHz stereo quality. It takes text prompts or image inputs, and it’s built for high-volume requests. Think sound effects, short loops, background audio previews, or anything where you need a quick, clean clip fast. This model launched first, in February 2026.
Lyria 3 Pro (lyria-3-pro-preview) generates up to 184 seconds of audio, roughly three minutes. It understands song structure, so you can prompt it with sections like intro, verse, chorus, and bridge, and it will produce a coherent composition rather than just an extended clip. It supports vocals, time-aligned lyrics, and full instrumental arrangements. This one arrived in March 2026, about a month after Clip.
Both are accessible via the Vertex AI API and Vertex AI Media Studio. Lyria 3 Pro is also available through Google AI Studio and the Gemini API.
What Makes This Different From the Consumer Version
Lyria 3 Pro has been rolling out in the Gemini app for paid subscribers, with daily generation limits tied to subscription tier (10 tracks per day on AI Plus, 20 on Pro, 50 on Ultra). That’s a consumer product with a quota ceiling.
The Vertex AI API path is different. It’s designed for production workloads, with broader regional availability than the Gemini consumer app. If you’re building an application that needs to generate bespoke audio at scale, whether that’s a gaming platform, a content tool, or a video production pipeline, the API route is the relevant one.
It’s also worth noting: Google trained Lyria 3 Pro on licensed and permissible data, and the company is offering IP indemnification for content generated using its models. If a third-party copyright claim arises from output you generated through the API, Google has committed to covering that. For any business building a commercial product on top of AI-generated audio, that’s a meaningful commitment that competitors like Suno and Udio, which have faced copyright lawsuits, currently cannot match.
Responsible AI Controls Built In
Every track generated by either model is embedded with SynthID watermarking, Google DeepMind’s imperceptible digital marker for identifying AI-generated content. Importantly, SynthID is designed to survive common audio edits like cropping, so provenance travels with the file even after post-production.
Both models also support C2PA metadata, the emerging content credentials standard that allows downstream platforms and tools to verify the origin of media.
On the content side, Lyria applies safety filters, recitation checks, and artist intent checks. If you prompt the model with a specific artist’s name, it treats that as broad stylistic inspiration rather than an instruction to imitate. Prompts that violate Google’s responsible AI guidelines are blocked before generation.
Vertex AI Now Covers Every Modality
The broader context here is that with music generation now in the mix, Vertex AI becomes the only platform with generative media models across video (Veo), images (Imagen), speech, and music, all accessible from a single API surface. That’s genuinely useful for production pipelines where you’re assembling multiple asset types from a text brief.
For a developer building, say, a short-form video tool, you can now go from a text prompt to a visual concept, to generated video footage, to a custom soundtrack, without leaving the Vertex AI ecosystem or stitching together tools from different vendors.
What You Should Know Before Building
A few practical things worth flagging:
- Both models are currently in pre-GA preview. Google’s “Pre-GA Offerings Terms” apply, meaning limited support is available and the product is offered as-is. That said, Google’s terms do permit use for production and commercial purposes during the preview period.
- Rate limits apply during preview, particularly for Lyria 3 Clip on high-volume requests.
- The structural prompting capabilities (verses, choruses, timed lyrics) are specific to Lyria 3 Pro. Lyria 3 Clip is optimised for short, fast generation and doesn’t carry those composition controls.
If you want to get started, the Vertex AI music generation guide is the right place to begin, with documentation covering both model identifiers, prompt structure, and API setup.
For teams that have been waiting for a production-ready, legally defensible path to AI music generation at scale, this preview is worth taking seriously.