Veo 3.1 API: you pay only for successful videos; Sume refunds

Google's Gemini API charges for Veo 3.1 only when the video is generated. Sume reserves cost on submit and refunds failures. What each means for your balance.

5 min readSume
All posts

Google's Gemini API pricing page (read 2026-10-02) says of Veo 3.1: "You will only be charged if your video is successfully generated", including when an audio processing issue stops a video from being produced. Sume works the other way round on timing: it reserves the estimated cost when you submit a job, captures it on success, and releases the reserve if the job fails or is cancelled.

What Google lists for Veo 3.1

The Gemini API pricing page lists three Veo 3.1 models, priced per second with audio, and shows no batch price for them.

Veo 3.1 on the Gemini API, per second (read 2026-10-02)
Model720p1080p4K
veo-3.1-generate-preview (Standard)$0.40$0.40$0.60
veo-3.1-fast-generate-preview (Fast)$0.10$0.12$0.30
veo-3.1-lite-generate-preview (Lite)$0.05$0.08Not supported

Charge on success versus hold then settle

On Google's side the bill appears only for a delivered video, so a failed call costs you nothing and there is no money held in between. On Sume the usage ledger has three states: reserved (estimated usage held before the provider runs), captured (billable usage taken after successful completion) and refunded (the hold released after failure or cancellation before capture).

The practical consequence is the balance you see while a job runs. A wallet with $10 in it that submits a job reserved at $1.25 shows less available until the job settles, and then either the capture sticks or the hold returns.

Where a hold can reject you

Because a hold must be covered up front, a balance that cannot cover the reserve gets 402 insufficient_credits, described in the error table as a balance that is not sufficient for the requested generation. That is a Sume behavior, not something Google's per-success billing does: top up before a large batch, and sum the reserves of your in-flight jobs, not only the price of one.

For images the Sume docs are plain that billing is all-or-nothing: a completed generation is billed in full, and a failed or cancelled one is not billed.

Which model on Sume

Sume's video catalog ids in the video docs do not include a Veo id. The Google video model there is gemini-omni-flash-1.1, which Google's page lists at $17.50 per million video output tokens, which it puts at about $0.10 per second for 720p. Price it from the catalog, multiply by 1.25 for the reserve, and use GET /v1/usage to see which rows were captured and which were refunded.

Sources

Related posts

More in Pricing

All Pricing posts

Written by Sume