Does Google keep Omni and Veo prompts for 55 days?
Google's Gemini API page says prompts, context and outputs are kept 55 days for abuse checks. Veo files last 2 days. How that differs from a Sume request.

Yes, for the Gemini API in general. Google's abuse-monitoring page says it retains prompts, context and outputs for 55 days to detect policy violations, and that flagged content can be reviewed by authorized employees under predefined guidelines. The page does not name Gemini Omni Flash or Veo 3.1, so treat it as the baseline for any Gemini API video call, not as a model-specific promise. This is read on 2026-10-08.
Three different clocks get mixed up in this question. The 55 days is the abuse log. Veo's 2-day figure is how long a generated video stays downloadable. And Omni's store setting only decides whether the next turn can edit the clip. Keeping them apart avoids a wrong answer in a security review.
Three retention numbers from Google's pages
The table lists each number with the page it comes from. None of them is a Sume number.
| Item | What the page says | Page |
|---|---|---|
| Abuse monitoring | Prompts, context and outputs kept 55 days | Abuse monitoring |
| Human review | Flagged content can be reviewed by authorized employees | Abuse monitoring |
| Enforcement | Rate limits, temporary suspension, or account closure for serious violations | Abuse monitoring |
| Veo 3.1 video files | Stored 2 days at most; extending a clip resets the timer | Veo 3.1 docs |
Omni store=false | The clip cannot be edited in a later turn with previous_interaction_id; no retention period is stated | Omni docs |
| Paid versus free | Pricing page: content is used to improve products on the free tier; paid users opt out | Pricing |
What the pages do not say
The Omni page gives no retention period for its stored interactions, and the abuse page does not say whether 55 days applies differently to video outputs. If a client needs an exact answer for a generated video, ask Google in writing rather than inferring it from these pages.
The SynthID watermark is a separate matter: Google's Omni docs say every generated video carries one, which is a provenance mark and not a retention setting.
How a Sume request differs
Sume does not list Veo 3.1. It lists Gemini Omni Flash 1.1 as gemini-omni-flash-1.1, and the model card in the repo describes it as a fal-served row. So Google's direct-API retention table above describes a request sent to Google, not a request sent to Sume.
On the Sume side, the video generation docs say video generation is ZDR-ineligible and that Sume has no zero-data-retention toggle. If your contract requires ZDR for video, that is a blocker on both routes, and you should read Sume's privacy material before you commit.
A short checklist before you send client footage
Decide what leaves your machine before you pick a model.
- Do not put signed URLs, customer lists or unreleased packaging into a prompt; assume a 55-day log on any Gemini API path.
- Download a Veo clip the same day. After 2 days the file is gone and extension needs a Veo-generated source.
- For Omni edits, keep
storeon only if you need multi-turn edits; otherwise the clip is not stored for that purpose. - On Sume, keep the job result from
GET /v1/jobs/:id/resultin your own storage if the clip is a deliverable.
What to do about it
Do not put anything in a prompt that you would not want retained for 55 days. That includes names, addresses and client details. If a project needs stricter handling, ask the vendor in writing before you use the model, since a retention window and a training-use rule are separate matters.
Sources
Related posts
More in Models
- Is GPT-6 Astra new this week? Sume has listed it since September
The OpenAI page for GPT-6 Astra states no release date. Sume's registry carries Astra with a rate card dated 2026-09-05, so it is not a new pick for Sume users.
- GPT-6 Astra on Sume holds a $10 reserve per turn: wallet check
Sume's billing code reserves 1000 cents on an Astra turn (2000 with Fast) and releases what the turn does not use. Check wallet headroom before you pick it.
- GPT Image 2.5 draft grid: four low-quality images for 4 cents
Four GPT Image 2.5 drafts in one Sume call at quality low cost 4 cents; a dollar buys 25 such calls. The arithmetic, the n field and when to go to high.
- Grok Imagine on Sume: 3 cents, one image per call, 9:20 phone ratios
Grok Imagine bills 3 cents per image on Sume, returns one image per call, lists 13 ratios including 9:20 and 19.5:9 phone shapes, and supports edits.
Written by Sume