Does Lyria 3.5 on the paid Gemini API train on my prompts?
Google's Gemini pricing page lists data use as No on the paid tier for Lyria models and Yes on the free tier. Lyria 3.5 has no free tier.

No for the paid tier, according to the Gemini API pricing page as read on 2026-10-04: its row for whether data is used to improve Google's products shows Yes on the free tier and No on the paid tier for the Lyria models. Lyria 3.5 is listed with no free tier at all, so a lyria-3.5 call is always a paid-tier call.
That is Google's statement about Google's API. This post covers what that row says, which Lyria models it applies to, and what to check when you reach Lyria through another provider.
What the pricing page shows
The rows below are taken from the pricing page, not from the music docs.
| Model | Price | Free tier | Data used to improve products |
|---|---|---|---|
| lyria-3.5 | $0.08 per song | None listed | Paid tier: No |
| lyria-3-clip-preview | $0.04 (30 s) | Listed, see page | Free: Yes, paid: No |
| lyria-3-pro-preview | $0.08 | Listed, see page | Free: Yes, paid: No |
Reading the row correctly
The row is about Google using prompts and outputs to improve its products. It does not say who owns the generated song or whether the audio can be used commercially; those questions are separate and the page does not answer them.
The Gemini music docs add that generated audio carries a SynthID watermark. That is a provenance mark, not a data-use setting.
- Free tier and paid tier are different terms. Check which one your key is on.
- The lyria-3-clip-preview and lyria-3-pro-preview models sit in a legacy family on the pricing page.
- lyria-3.5 is the only one with no free tier.
Using Lyria 3.5 through Sume
Sume's Music Router can route to lyria-3.5, lyria-3-pro or the default sume/music-auto. The price is a fixed $0.125 per generation, and job.request.routed_model tells you which engine ran. Sume's own data-handling terms are what apply to your Sume account, so read them and the Sume docs for the router itself rather than assuming Google's paid-tier row transfers.
curl -X POST https://api.sume.com/v1/music-router/generate \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: music-data-use-001" \
-d '{
"model": "sume/music-auto",
"prompt": "Warm lo-fi hip hop, 84 BPM, C minor. Dusty Rhodes chords, brushed drums. A 30-second track. Instrumental, no vocals."
}'Quick checklist
- Confirm the tier your Gemini key is on before sending anything sensitive.
- Keep brand-confidential lyrics out of any free-tier prompt.
- Record the engine that ran: routed_model on a Sume job.
- Treat ownership and licensing as a separate question from data use.
Sources
Related posts
More in Pricing
- MiniMax H3: 2K by default there, 768p by default on Sume
MiniMax's page says H3 defaults to 2K; Sume's default is 768p, and 2K bills $0.1625 per second versus $0.075. Per-second and 10 s costs by resolution.
- MiniMax H3 Max price per second on Sume: 480p, 768p, 1080p vs Wan
H3 Max lists $0.05, $0.08 and $0.16 a second at 480p, 768p and 1080p. On Sume that is $0.0625, $0.10 and $0.20. Compared with Wan 3.0 at each step.
- Four edit variants in one call: n limits and cost per Sume image model
Most Sume image models accept n up to 4, Grok Imagine only 1. You pay cost_usd times n, so four variants of an edit cost four edits.
- Nano Banana 2 API has no free tier: size a 20-image test
Google lists no free tier for Nano Banana 2 image output. Work out a 20-image test budget by tier, then read the live Sume price from the endpoints call.
Written by Sume