Every Veo endpoint on the Gemini API shuts down on 22 October 2026 — veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview, all on the same day. There is no downgrade: Veo 3.0 and 2.0 went on 30 June 2026, and no generally available Veo has ever existed on this API. Google names one replacement, gemini-omni-1.1-flash. Priced per second of finished 720p video, that move is a 75% cut from Veo 3.1, roughly flat from Fast, and a 103% increase from Lite. The cheapest endpoint dies hardest. This is an API-only deadline — the Gemini app, Flow and the free consumer allowance are unaffected.
The deadline, and why there is nothing to pin to
Most model retirements give you an escape hatch: stay on the previous version, buy a quarter, migrate properly later. This one does not, and the reason is structural rather than punitive.
Google’s Gemini API deprecation table lists the three Veo 3.1 variants like this:
| Model | Released | Shutdown | Replacement |
|---|---|---|---|
veo-3.1-generate-preview | 15 Oct 2025 | 22 Oct 2026 | gemini-omni-1.1-flash |
veo-3.1-fast-generate-preview | 15 Oct 2025 | 22 Oct 2026 | gemini-omni-1.1-flash |
veo-3.1-lite-generate-preview | 31 Mar 2026 | 22 Oct 2026 | gemini-omni-1.1-flash |
veo-3.0-generate-001 | 9 Sep 2025 | 30 Jun 2026 (gone) | — |
veo-3.0-fast-generate-001 | 9 Sep 2025 | 30 Jun 2026 (gone) | — |
veo-2.0-generate-001 | 9 Apr 2025 | 30 Jun 2026 (gone) | — |
Read the suffixes. Every one of them ends in -preview. Google has served Veo on the Gemini API for eighteen months and never graduated it to GA. That means the normal advice — “pin the last stable version and move on your own schedule” — has no target. After 22 October, the count of Veo models callable on the Gemini API is zero.
The third row deserves its own sentence. veo-3.1-lite-generate-preview shipped on 31 March 2026 and is switched off on 22 October 2026: a 205-day life. It is also the cheapest video-with-audio endpoint Google has ever sold. Teams that adopted it specifically because it was cheap are being given seven months and a bill that doubles.
One mitigation, and it cuts both ways: Google’s own page describes these as “the earliest possible dates on which a model might be retired.” A slip is possible. Treat it as upside, not as a plan — the same reasoning applies here as to the Gemini 2.5 migration, where an undated arrangement turned out to be the more dangerous one.
What the replacement actually costs
Here is the part no vendor page does, because it requires converting two different billing models into the same unit.
Veo 3.1 billed per second of output video. Google’s rate card, video with audio:
| Variant | 720p | 1080p | 4K |
|---|---|---|---|
| Veo 3.1 | $0.40/sec | $0.40/sec | $0.60/sec |
| Veo 3.1 Fast | $0.10/sec | $0.12/sec | $0.30/sec |
| Veo 3.1 Lite | $0.05/sec | $0.08/sec | — |
gemini-omni-1.1-flash bills per token. Video output is $17.50 per million tokens, and Google supplies the conversion: “Billing is based on total output token consumption, calculated at a rate of 5,792 tokens per second of 720p video. Under Standard pricing, this equates to an effective price of approximately $0.10 per second.”
Do the multiplication precisely and 5,792 × $17.50 ÷ 1,000,000 = $0.10136 per second at 720p. Now the three migrations land in three different places:
| You are on | Per sec | Omni 1.1 per sec | Change |
|---|---|---|---|
| Veo 3.1 (720p) | $0.40 | $0.1014 | −75% |
| Veo 3.1 Fast (720p) | $0.10 | $0.1014 | +1.4% |
| Veo 3.1 Lite (720p) | $0.05 | $0.1014 | +103% |
Sized on a real workload — 500 clips a month at 8 seconds each, 720p with audio, 4,000 seconds of finished video:
| Model | Per second | 8-sec clip | Monthly |
|---|---|---|---|
| Veo 3.1 | $0.40 | $3.20 | $1,600.00 |
| Veo 3.1 Fast | $0.10 | $0.80 | $400.00 |
| Veo 3.1 Lite | $0.05 | $0.40 | $200.00 |
| gemini-omni-1.1-flash | $0.1014 | $0.81 | $405.44 |
| Runway Gen-4.5 (API) | $0.12 | $0.96 | $480.00 |
| Kling 3.0 (720p + audio) | $0.126 | $1.01 | $504.00 |
Prices verified 8 October 2026 against the Gemini API pricing page, Runway’s API pricing and Kling’s published developer rates. Our running record of these moves is in the pricing tracker and the model retirement tracker.
Two things fall out of that table that nobody is saying out loud.
The replacement is priced almost exactly at Veo 3.1 Fast. Within 1.4%. If Fast was your production endpoint, this migration is a code change and a re-eval, not a budget conversation.
The audio surcharge disappeared. Veo charged for audio: Runway’s resale card, which carries both, lists Veo 3.1 at 40 credits per second with audio and 20 without — a 2x premium Google’s own rate card never surfaces, since it publishes only the with-audio default. gemini-omni-1.1-flash has a single video rate. So silent work migrates worse than audio work: a team generating muted B-roll on Veo 3.1 at an effective $0.20/sec sees a 49% cut, while a team generating dialogue at $0.40/sec sees 75%.
The resolution bill Google does not publish
This is the live gap in the rate card, and it will surprise someone’s finance team.
Google publishes one conversion: 5,792 tokens per second of 720p. The gemini-omni-1.1-flash documentation lists four supported output resolutions — “360p, 720p (default), 1080p (upscaled), 4k (upscaled)” — and the pricing page gives a token rate for none of the other three.
Runway resells the same model and does break it out:
| Resolution | Runway credits/sec | At $0.01/credit | vs. its own 720p |
|---|---|---|---|
| 360p | 3.4 | $0.034 | 0.34x |
| 720p | 10 | $0.10 | 1.0x |
| 1080p | 15 | $0.15 | 1.5x |
| 4K | 30 | $0.30 | 3.0x |
Runway’s 720p rate lands on Google’s published ~$0.10, and its Veo 3.1 with-audio rate ($0.40) lands on Google’s $0.40 — so on the lines where both publish a comparable number, the resale is at par. That makes the 1.5x and 3x steps a defensible planning assumption. It is still a third party’s card, not Google’s list; measure your first real bill before you commit a budget.
Three consequences worth planning around:
- 1080p stopped being free. On Veo 3.1 standard, 720p and 1080p cost the same $0.40 — so the rational choice was always 1080p. On the replacement, 1080p is a 1.5x surcharge. Anyone who hardcoded 1080p because it cost nothing extra is about to start paying for it.
- 1080p and 4K are upscaled, not natively rendered. That is Google’s own word. Veo 3.1 priced 4K as a separate, higher-cost render ($0.60/sec). You are now paying roughly 3x the 720p rate for an upscale. Check that the output still clears your quality bar before you assume the 4K migration is like-for-like.
- There is a new input charge. Veo billed output seconds and nothing else.
gemini-omni-1.1-flashbills input at $1.50 per million tokens across text, image, video and audio. Image-to-video start frames, reference images and clips you upload for extension are now a billable line that did not exist. Runway prices the same thing concretely at 1 credit per reference image and 1 credit per second of input video — small per call, real at volume.
What you gain, and one place Google’s docs disagree
The migration is not pure loss, and the feature story is better than the deprecation notice suggests.
gemini-omni-1.1-flash generates video with synchronised audio, supports image-to-video with reference images, interpolates between first and last frames, and extends clips — up to about 40 seconds of total length, with input clips for editing capped at 10 seconds. It also accepts text, image, audio and video as simultaneous inputs, which Veo did not.
But Google’s own documentation contradicts itself on exactly the features a migrating team will check first. The Gemini API video overview tells you to “Use Veo 3.1 for specific capabilities like scene extension, last-frame control, or integration with legacy pipelines” — while the gemini-omni-1.1-flash documentation describes scene extension (“Extend an existing video by generating a seamless continuation at the tail end of the clip”) and frame interpolation as supported features of the replacement.
Both pages are Google’s. One of them is stale. Given that the model it points you toward is switched off in a fortnight, assume the Omni docs are current — but test scene extension and last-frame control against your actual prompts before 22 October, because that overview page is the only published warning that the replacement might not cover them, and you have no fallback if it is right.
If you are leaving Google
For Veo 3.1 Lite users facing a doubling, shopping the move is rational rather than defeatist. Two serious non-Google exits, both of which generate native audio:
- Runway Gen-4.5 — $0.12/sec on the API (12 credits), the strongest cinematic quality in the category, and the only one of the three with an integrated editor (Aleph) and performance capture (Act-Two). It is 18% more than Omni at 720p. Subscription plans are separate from API credits: Free, Standard $15/mo ($12 annual), Pro $35 ($28), Max $95 ($76). Our Runway vs Veo 3 comparison covers the quality axis in detail.
- Kling 3.0 — $0.084/sec at 720p silent, $0.126 with native audio, $0.168 at 1080p with audio, and 4K from $0.42. Kling is the only option here that is cheaper than the replacement if you do not need audio, and it generates far longer single clips. The trade-offs are the ones in our Kling review: Chinese content restrictions and data-residency questions that disqualify it for some buyers outright.
If you are on Veo 3.1 standard, the honest answer is that the replacement is the cheapest credible option on this list and switching vendors would cost you money as well as engineering time. Take the 75% cut. For the wider field, see best AI video tools; for where Veo sits now, the Veo 3 review and the Gemini review.
Migration checklist
- Grep for
veo-3.1across everything — application code, notebooks, cron jobs, Terraform, CI config, SDK defaults. Model strings outlive the code that owned them. - Identify which variant each call site uses. The cost direction is opposite for Lite and standard; a single blended estimate will be wrong for both.
- Re-price at your actual resolution, not at 720p. If you hardcoded 1080p because it was free on standard Veo, that is now a 1.5x line.
- Budget the new input charge. Count reference images and uploaded clips per call and multiply; this line did not exist before.
- Test scene extension and last-frame control explicitly, because Google’s two docs pages disagree about whether the replacement has them.
- Check 4K output quality if you ship 4K — you are moving from a native render to an upscale.
- Confirm your Vertex position separately if you also call Veo there. Vertex runs its own deprecation calendar and the Gemini API date does not automatically apply.
- Do it before 22 October. There is no older Veo to fall back to, and a failed video endpoint is a failed render queue, not a degraded response.
What changed
- 8 Oct 2026: Page created. All three Veo 3.1 variants confirmed for 22 October shutdown against Google’s deprecation table;
gemini-omni-1.1-flashpriced at $0.10136/sec at 720p from Google’s own 5,792-tokens-per-second conversion. - 8 Oct 2026: Google publishes no token rate above 720p for the replacement despite documenting 360p, 1080p and 4K as supported; Runway’s at-par resale card used as the stated proxy (1.5x and 3x).
- 27 Aug 2026:
gemini-omni-1.1-flashreleased, with no shutdown date announced. - 30 Jun 2026: Veo 3.0, Veo 3.0 Fast and Veo 2.0 shut down on the Gemini API, leaving Veo 3.1 Preview as the only Veo generation.
- 31 Mar 2026:
veo-3.1-lite-generate-previewreleased — 205 days before its own shutdown date. - 19 May 2026: Gemini Omni announced at Google I/O as a multimodal video series; at the time it appeared to sit alongside Veo rather than replace it.
Frequently asked questions
When exactly does Veo 3.1 shut down?
22 October 2026, for all three variants on the Gemini API. Google's deprecation table lists veo-3.1-generate-preview (released 15 October 2025), veo-3.1-fast-generate-preview (15 October 2025) and veo-3.1-lite-generate-preview (31 March 2026) with the same shutdown date and the same recommended replacement, gemini-omni-1.1-flash. The older GA models — veo-3.0-generate-001, veo-3.0-fast-generate-001 and veo-2.0-generate-001 — were already shut down on 30 June 2026. One caveat in Google's favour: the page describes its shutdown dates as 'the earliest possible dates on which a model might be retired', so 22 October is a floor rather than a promise. Plan for the floor.
Can I just pin to an older Veo version?
No, and this is what makes the deadline harder than the Gemini 2.5 access gate. Every Veo model Google has ever served on the Gemini API carries a -preview suffix; there has never been a generally available Veo endpoint. The 3.0 and 2.0 generations are already gone as of 30 June 2026. So after 22 October there is no Veo model of any version on the Gemini API, and no downgrade path. This is the unusual case where the entire product line leaves the API at once.
Is the replacement cheaper or more expensive?
It depends entirely on which variant you were on, and the direction is not the same for everyone. At 720p, gemini-omni-1.1-flash works out at about $0.101 per second of generated video. Veo 3.1 was $0.40 per second, Veo 3.1 Fast $0.10, and Veo 3.1 Lite $0.05. So the move is a 75% cut from the standard model, essentially flat from Fast, and a 103% increase from Lite. The teams facing a doubling are the ones who had already optimised for cost, which is the opposite of how migrations usually land.
Does this affect the Gemini app, Flow, or the 10 free videos a month?
No. The shutdown is on the Gemini API only. Google lists Veo 3.1 as available in the Gemini API and Vertex AI, and separately 'in the Gemini app and Flow', plus Vids and Photos. Those consumer and creative surfaces have their own release schedules and none of them appears on the Gemini API deprecation table. Vertex AI also runs a separate deprecation calendar from the Gemini API; this analysis could not reach Google's Vertex deprecation page to confirm a date either way, so if Vertex is your escape hatch, confirm it with your account team rather than assuming the Gemini API date applies.
What does gemini-omni-1.1-flash cost at 1080p or 4K?
Google does not say. Its pricing page gives exactly one conversion — 5,792 output tokens per second of 720p video — and publishes no token rate for 360p, 1080p or 4K, even though the model documents all four as supported resolutions. The best available proxy is Runway's resale card, which lists the same model at 3.4 credits per second for 360p, 10 for 720p, 15 for 1080p and 30 for 4K. At $0.01 a credit its 720p rate lands on Google's published $0.10, so the 1.5x and 3x steps to 1080p and 4K are a reasonable planning assumption — but they are a third party's rate card, not Google's list price. Measure your own first bill.