What is Lyria 3.5?
Lyria 3.5 is Google DeepMind's music generation model and the successor to Lyria 3 Pro. It writes complete songs with vocals, timed lyrics and arrangements, structured into verses, choruses and bridges, from a text prompt or from up to 10 images. Output is 44.1 kHz stereo MP3 by default, with WAV available.
It reached Google Flow Music on 29 July 2026 and became generally available in the Gemini API on 3 September 2026, followed by the Gemini app for all users, Google AI Studio and Google Vids. The earlier Lyria 3 Clip (30-second clips) and Lyria 3 Pro preview models, launched on 25 March 2026, remain on the API and are now labelled legacy.
The API charges per track: $0.08 for a full Lyria 3.5 or Lyria 3 Pro song and $0.04 for a Lyria 3 Clip. There is no free API tier. Every track carries a SynthID audio watermark.
Is Lyria 3.5 worth it?
Lyria 3.5 matters less for how it sounds than for how it is sold: it is one of very few leading music models with a plain per-song API price. At $0.08 a track, 1,000 full songs cost $80 through the Gemini API, with no subscription, credit pack or download cap. Suno and Udio, the two consumer leaders, sell credits through their own apps, and neither publishes a general developer API.
The model writes whole songs, not loops. According to the docs it handles verses, choruses and bridges, writes and sings its own lyrics or yours, and runs for "a couple of minutes", with length set in the prompt rather than by a parameter. It takes up to 10 images alongside the text, so a product photo or a storyboard can set the mood. Output is 44.1 kHz stereo, MP3 by default or WAV on request. The Lyria 3 Clip preview, at $0.04 for exactly 30 seconds, is the cheap way to test prompts first.
The constraints are firm. Generation is single-turn: you cannot ask it to change the second chorus, only to start again, and results vary between calls with the same prompt. Safety filters block prompts that name an artist's voice or ask for copyrighted lyrics. There is no batch API, no caching and no function calling.
Rights are the open question. Google does not say what Lyria was trained on, and every output carries an inaudible SynthID watermark that identifies it as AI-made. On the paid API tier, Google states prompts are not used to improve its products.
Our take: for an app that needs background music, jingles or game themes on demand, the flat $0.08 per track makes Lyria 3.5 the easiest choice to budget; for a musician who wants to shape a song section by section, Suno's editor still does more.
Where it earns its place
A mobile game studio with 40 levels to score
Generate a distinct theme for every level without hiring a composer
Draft prompts on Lyria 3 Clip at $0.04 each, then render finals with Lyria 3.5; 40 finals with five takes each cost 200 x $0.08, or $16, before anyone listens.
A video platform with a self-serve editor
Offer users a custom backing track that matches their footage
Pass up to 10 frames from the clip with a mood prompt; at $0.08 a song, 50,000 tracks a month cost $4,000, and the SynthID watermark lets the platform label the audio as AI-made.
A marketing agency producing radio and social ads
Turn a brief into a jingle with sung lyrics
Write the lyrics into the prompt with section tags such as [Verse] and [Chorus]; because a song cannot be edited, plan on several full takes per brief and pick one.
Before you commit
- Setup
- Create a paid Gemini API key in Google AI Studio and call lyria-3.5 through the Interactions API; the response returns base64 audio and the generated lyrics.
- Integrations
- Gemini API Interactions API with Python, JavaScript, Go and REST examples; Lyria RealTime for streaming generation is a separate model.
- Data and privacy
- On the paid tier Google states prompts and outputs are not used to improve its products; all audio carries a SynthID watermark.
- Cost at scale
- Flat $0.08 per full song regardless of length, so 10,000 songs cost $800 and 100,000 cost $8,000.
- Limits
- Input limit of 131,072 tokens; single-turn generation with no editing; no batch, caching or structured output; prompts requesting artist voices or copyrighted lyrics are blocked.
What’s good
- $0.08 per full song through the Gemini API, no subscription
- Full songs with vocals, lyrics and structure, from text or up to 10 images
- 44.1 kHz stereo, MP3 or WAV
- 30-second Clip model at $0.04 for testing prompts
- Paid-tier prompts are not used to improve Google's products
What’s not
- Single-turn only: no editing a generated song
- Length is steered by prompt, not set exactly
- No free API tier, no batch API
- Training data not disclosed
- Blocks prompts naming artists' voices
Best for
Developers who need full-length songs with vocals generated inside their own app at a predictable per-track price.
Lyria 3.5: common questions
- Is Lyria 3.5 worth it?
- Lyria 3.5 is the music model to call from code if you need full songs with vocals at a flat price, though it gives you one take per prompt and no way to edit it. We rate it 4.2 out of 5.
- How much does Lyria 3.5 cost?
- Gemini API paid tier only: Lyria 3.5 $0.08 per song; legacy Lyria 3 Pro Preview $0.08 per song and Lyria 3 Clip Preview $0.04 per 30-second clip. No free API tier. Also in the Gemini app and Flow Music on Google's consumer plans.
- What are the downsides of Lyria 3.5?
- Single-turn only: no editing a generated song; length is steered by prompt, not set exactly; no free API tier, no batch API; training data not disclosed; blocks prompts naming artists' voices.
- Who is Lyria 3.5 best for?
- Developers who need full-length songs with vocals generated inside their own app at a predictable per-track price.
- How much does Lyria 3.5 cost for a team or in production?
- Flat $0.08 per full song regardless of length, so 10,000 songs cost $800 and 100,000 cost $8,000.
- Is Lyria 3.5 safe for company data?
- On the paid tier Google states prompts and outputs are not used to improve its products; all audio carries a SynthID watermark.
Answers checked Nov 6, 2026.
Make Lyria 3.5? Show this rating on your site
