If your pipeline calls gemini-omni-1.1-flash, it starts getting video from a different model on December 3, billed at the new model's rates, with no code change on your side. Google made Gemini Omni 1.2 Flash generally available on Google Cloud on November 5 at $0.07 per second of 720p video, which it describes as 30% below Omni 1.1 Flash. The cut is real, and you get it without doing anything. You still have 25 days to retest approved creative, re-cost your upscales, and get residency and indemnity answers that neither launch post gives.
Omni 1.2 Flash is Google's video generation and editing model. Google's launch post presents it as the first generally available Omni, although the Gemini API changelog already called 1.1 Flash GA on August 27. On Agent Platform, 1.1 Flash is still labelled preview, which matters to anyone whose procurement team refuses preview software.
What Google Shipped on November 5
Omni 1.2 Flash extends clips to 15 seconds and offers Provisioned Throughput, with drafts at 360p or 720p and upscaling to 1080p or 4K. Google's launch post sets the duration range at 3 to 15 seconds in one-second steps, up from a 10-second ceiling on 1.1. The model card adds the working limits: 16:9 and 9:16 only, up to three videos per prompt, and up to 10 images per prompt. The Gemini API changelog also records doubled paid-tier rate limits.
The evidence for quality is thin. Google publishes side-by-side examples and customer quotes, with no benchmark of its own, and The Decoder's write-up attributes the improvements to Google's announcement. The most useful line for a buyer comes from Photoroom's CEO, quoted in Google's Cloud post: "With general availability and SLAs, we can rely on it in production for our sellers and enterprise customers." Google also says it is still working on fast sports action, rapid script changes and complex physics, so test those shots first if your brand depends on them.
The December 3 Routing Covers One Model ID
The December 3 switch applies to the Gemini API ID, and Google Cloud customers are on a different clock. The Gemini API pricing page says requests to gemini-omni-1.1-flash will route automatically to gemini-omni-1.2-flash and bill at 1.2 rates from that date. Neither that notice nor the changelog entry mentions an opt-out. Google describes this as automatic routing, so expect the call to return video from 1.2 and your error monitoring to stay quiet.
The changelog notice is dated November 5, which leaves 28 days of warning. Omni 1.1 Flash itself was released on August 27, so it gets 98 days before its traffic moves.
On Gemini Enterprise Agent Platform, the older model has a different ID. The Omni 1.1 Flash model card lists gemini-omni-1.1-flash-preview, labels it deprecated, recommends 1.2, and gives no date. Because it is a preview model, Google's service terms let it be "changed, suspended or discontinued at any time without prior notice," outside any SLA or indemnity.
The new model comes with a firmer promise. Google's model lifecycle page lists gemini-omni-1.2-flash with a retirement date of "November 5, 2027 or later," and says listed dates will not move earlier. The Gemini API's deprecations page lists no shutdown date for 1.2.
What the Price Cut Is Worth After Retakes and Upscales
The 30% cut applies per second, so a team that moves to full 15-second clips will spend about the same per clip. At Google's published token rates, a 10-second 720p clip on 1.1 cost about $1.01 ($17.50 per million video tokens, 5,792 tokens per second). A 15-second clip on 1.2 costs $1.04 at $12.00 per million. Longer takes are useful, but your budget only falls if clip length stays put.
The other number to re-cost is the upscale. Google lists about $0.104 per second for 1080p and $0.209 for 4K, and the upscaling documentation shows how those are built. Upscaling is a separate request, billed at 2,896 output tokens per second from 720p to 1080p and 11,584 from 720p to 4K. Add those to the 5,792 for generating at 720p and you land on the 8,688 and 17,376 tokens per second on the price list. So the published 4K rate is the cost of one draft plus one upscale, and any retake you throw away costs extra.
Here is one finished 15-second clip under the four-takes-per-shot assumption from our October cost comparison. The arithmetic is ours, using Google's token rates and video output tokens only.
| Workflow for one 15-second final | Video tokens billed | Cost at $12 per 1M |
|---|---|---|
| One 720p take, used as is | 86,880 | $1.04 |
| Four 360p takes, winner upscaled to 1080p | 202,740 | $2.43 |
| Four 720p takes, winner upscaled to 1080p | 390,960 | $4.69 |
| Four 720p takes, winner upscaled to 4K | 521,280 | $6.26 |
| Four takes all rendered through to 4K | 1,042,560 | $12.51 |
Prices are from Google's Gemini API page, checked November 8, 2026. Prompts and reference assets bill separately as input at $1.50 per million tokens, and thinking tokens bill at the $9.00 text output rate, while standalone upscaling carries zero input tokens. The gap between the top and bottom rows is a process decision: draft cheap, approve, then upscale once. Direct 360p-to-4K upscaling is not supported, so an upscaled 4K deliverable needs a 720p master. Treat every conversational edit as another billed render until your own logs say otherwise.
Gemini Enterprise Moved From Veo 3.1 to Omni
Business users on Gemini Enterprise have already changed models. The Gemini Enterprise release notes for November 6 say Omni 1.2 Flash "replaces Veo 3.1 for video generation in the global region," and that an administrator must turn on the "Enable video generation" setting before users can create video. The entry says nothing about apps in other regions.
Google's own messaging on the business surface conflicts. The launch post says the model is "generally available across the Gemini API and Gemini for business," while the Cloud post says "coming soon to Gemini for business." Check your own tenant before you tell a marketing team what they have.
If your creative team built prompt templates on Veo 3.1 in Gemini Enterprise, those templates now run on another model. Brand review signed off on Veo's output, so that sign-off no longer covers what users generate.
Omni Runs Only on the Global Endpoint
The Omni 1.2 Flash model card lists global as its only location for both pay-as-you-go and Provisioned Throughput. Google's data residency page says requests to a global endpoint "may be processed in any Google Cloud location around the world" and that such endpoints don't provide data residency guarantees. Data at rest stays in the location you choose, so the exposure is processing. Reference photos of unreleased products and of real people are the inputs to worry about.
The Veo 3.1 model card lists us-central1. A US team with a US-only processing rule therefore loses something by moving from Veo to Omni. Neither model card lists an EU location, so a European team with strict residency requirements has no in-region option on either one.
Organization policy can block the switch outright. The November 6 Gemini Enterprise notes say that if your resource location policy excludes global, administrators can't turn on Gemini models that aren't supported in the app's location. That entry doesn't name Omni, so test the toggle in your tenant.
On training, Google's zero data retention page says it won't use your data to train models without permission, but the page does not mention video generation. Google's responsible AI guidance for video, which documents the celebrity and child safety filters, covers Veo and doesn't mention Omni either. You will need to learn Omni's filter behaviour by testing it.
Get the Indemnity Answer in Writing
Neither launch post mentions IP indemnity, and that is the term a brand team should care about most. When we compared video vendors in October, Google's April indemnified-services list named generally available versions of models including Veo on the Agent Platform API. We could not read the current list for this article, so we can't tell you whether Omni 1.2 Flash is on it. GA status removes one obstacle, since the service terms exclude pre-GA offerings from indemnity.
The same terms limit what any covered output gets. The generated-output protection applies to "unmodified" output and excludes trademark claims arising from use "in trade or commerce." Those two conditions bite harder on Omni than on a generate-only model, because multi-turn editing, packaging typography and brand colours are what Google is selling here. Ask your account team whether a clip revised through conversational editing counts as unmodified, and whether a standalone upscale changes the answer. Our image indemnity comparison covers the same trademark gap for stills.
Outputs carry a SynthID watermark, and the model card lists Content Credentials (C2PA) as supported.
Size Provisioned Throughput From Measured Tokens
Video consumes reserved capacity quickly, so size any purchase from tokens per finished clip. Google's November 5 release note puts Omni 1.2 Flash at 685 tokens per second per GSU, and the supported models page counts each output video token as eight against that allowance. By our arithmetic, sustaining one second of 720p output per wall-clock second takes about 68 GSUs, and one GSU clears a 15-second 720p clip about every 17 minutes, or roughly 3.5 an hour, before input and reasoning tokens.
The Omni throughput documentation sets enforcement windows from 2,000 seconds at 1 to 9 GSUs down to 30 seconds at 67 or more. For 4K clips of 10 seconds or longer, it tells you to compare the estimator's result against its 4K table and order the larger number. Our break-even analysis of reserved capacity covers when a commitment pays off at all.
The Case for Keeping Veo 3.1 a While Longer
Staying on Veo 3.1 is defensible for some workloads through the spring. On Agent Platform, Veo 3.1 and Veo 3.1 Fast are GA with retirement dates of "May 4, 2027 or later", though Google already names Omni 1.2 Flash as the recommended upgrade. Veo offers a regional US endpoint, and its GA versions date to November 2025.
The argument against staying is price. Runway's API price list showed Veo 3.1 at $0.40 per second with audio and Veo 3.1 Fast at $0.15 when checked November 8, 2026. That is a reseller's rate and doesn't state a resolution, so treat the comparison as directional. Runway lists gemini_omni_flash_1.2 at a base $0.07 per second for 720p and an unversioned gemini_omni_flash at $0.10, so if you buy Omni through a reseller, check which version your calls use.
What to Do Before December 3
This Week:
- Search your code, workflow tools and vendor configs for
gemini-omni-1.1-flashandgemini-omni-1.1-flash-preview, and record which platform each call uses. - Have your Gemini Enterprise administrator confirm the state of "Enable video generation," and whether your resource location policy allows
global. - Ask your Google account team, in writing, whether Omni 1.2 Flash is on the indemnified-services list and how edited and upscaled clips are treated.
Before December 3:
- Pick 20 approved prompts that represent your hardest brand work (packaging text, faces, product colours) and run them on 1.1 and 1.2 side by side while both are available. Have the person who signs off creative score them.
- Pin to
gemini-omni-1.2-flashexplicitly once it passes, so your logs and your model ID agree. - Set the
seedand billinglabelsthat the Agent Platform release note added for Interactions API requests, so you can reproduce a run and trace its spend per campaign. - Change the workflow to draft at 360p or 720p and upscale only the approved take, and rebuild your per-clip budget using the table above.
Before Your Next Renewal or Reservation:
- Decide whether global processing is acceptable for reference assets. If it isn't, keep those jobs on Veo 3.1 in
us-central1and put the May 2027 date on the roadmap. - Size any Provisioned Throughput from measured tokens per finished clip, retakes included.
The Bottom Line
Video models are turning over faster than creative approval cycles. Veo 3.1's preview IDs carry an October 22, 2026 shutdown date on the Gemini API, Omni 1.1 lasts 98 days there before rerouting, and OpenAI's Sora shutdown earlier this year left production users looking for somewhere else to go. Our review of model deprecation windows compares the notice each platform gives. For a team planning a year of production video, Omni 1.2 Flash's 12-month minimum on Agent Platform is the commitment to build around.
Run the regression set before December 3, and record which model produced every approved clip.
Continue Reading
- Your 30-Second AI Video Spot Can Cost $19 to $86 in API Fees
- Adobe, Google and Getty Offer Different AI Image Indemnities
- Model Deprecation: Bedrock Keeps Claude 4 Months Past Anthropic
- Annual Azure and Google AI reservations break even at 84% and 75%
- Meta's Commercial Terms Put All AI Ad Risks on Advertisers
- Google's Antigravity Lacks SOC 2 and ISO 27001 as Code Assist Ends