Musicians and hobbyists can try full three-minute tracks free in Flow Music; the API is paid only and UK availability is unconfirmed.
Kenning's take
What it does
Lyria 3.5 is Google DeepMind's latest music generation model. Google says it improves musicality, lyrics and vocal quality, and lets users set tempo and song length, from a short clip up to a full three-minute track. It launched in Google Flow Music on 29 July 2026 and is listed in the Gemini API as lyria-3.5.
- Full songs up to 3 minutes
- More realistic vocals with better pronunciation
- Lyrics that follow song structure
- Tempo and duration control
- Image-to-music
- SynthID watermark on every track
Who it suits. Musicians and creators making full AI-generated songs, and developers adding music generation to apps.
Key facts
- Category
- AI models
- Part of
- Google Flow Music
- Made by
- Google DeepMind
- Based in
- United States
- Released
- 29 July 2026
- How to get it
- in Google Flow Music (free and paid plans) and in the Gemini API as lyria-3.5 (paid tier only)
- Works on
- Web, API
- Free plan
- Google Flow Music free plan with daily top-up credits
- API price
- $0.08 per song (full song); not on the API free tier
Pricing
| Plan | Price | Billed |
|---|---|---|
| Free | $0 | month |
| Starter | $8 | month |
| Plus | $24 | month |
| Member | $64 | month |
Reviews and write-ups
- RouteNote Blog 2026-08-04
FAQ
When was Lyria 3.5 released?
It launched in Google Flow Music on 29 July 2026.
How long can Lyria 3.5 songs be?
Up to 3 minutes.
How much does Lyria 3.5 cost?
Google Flow Music has a free plan and paid plans from $8 a month. In the Gemini API it costs $0.08 per full song.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.