On this page · 10 sections
Summary. From 3 December 2026, every Gemini API request that names gemini-omni-1.1-flash is served by gemini-omni-1.2-flash and billed at the 1.2 rates, with no code change needed. That is a price cut, not a rise. Video output on 1.2 Flash costs $12.00 per million tokens against $17.50 on 1.1, and Google counts 5,792 tokens for each second of 720p video, so a second of 720p falls from about $0.101 to about $0.070, a drop of 31.4%. A 10-second clip goes from about $1.01 to $0.70, and 1.2 can now make 15-second clips for about $1.04. Input stays at $1.50 per million tokens, and neither model has a free tier. Google released 1.2 Flash as generally available on 5 November 2026 and calls the 1.1 period a 4-week transition window. The catch is behaviour, not price: 1.2 lists native audio and longer clips that 1.1 did not, and the Tier 1 rate limit is 40 requests a day. Test your prompts on 1.2 before 3 December rather than after.
About this article. For product, engineering and marketing-technology teams that call Gemini Omni Flash through the Gemini API, and for finance leads who budget AI video. We checked every price, model ID, limit and date on 17 November 2026 against Google's own pricing page, model page, deprecations table, release notes, developer guide, terms and launch blog post, plus Google Cloud's release notes. Google can change prices and dates; check the pricing page before you commit a budget. This is not financial advice, and no vendor paid for inclusion. Researched with AI assistance and checked against the primary sources listed in the references.
What Google announced, and when
Google's Gemini API release notes for 5 November 2026 carry two linked entries. The first makes gemini-omni-1.2-flash generally available, with clips of up to 15 seconds, better consistency across up to 6 reference images, video output pricing 30% lower at $12.00 per million tokens, and doubled rate limits on the paid tiers. The second is the routing notice: from 3 December 2026, calls to gemini-omni-1.1-flash go to 1.2 and are charged at 1.2 rates.
The deprecations table gives the dates in one place. Omni 1.1 Flash was released on 27 August 2026 and has 3 December 2026 in its shutdown column, with 1.2 Flash as the replacement and a note that the endpoint routes requests automatically from that date. Omni 1.2 Flash was released on 5 November 2026 and has no shutdown date announced. The model page shows both IDs as stable and the original gemini-omni-flash-preview as shut down.
Google's launch post, published on 5 November, frames the release around the same 30% cut and gives the headline figure of $0.07 per second for 720p video. The dates below come from the deprecations table and the video generation guide.
| Date | Event | What it means for you |
|---|---|---|
| 22 October 2026 | Veo 3.1 Preview models shut down in the Gemini API | Their named replacement is Omni 1.2 Flash |
| 5 November 2026 | gemini-omni-1.2-flash generally available | Lower price, 15-second clips, higher limits |
| 3 December 2026 | 1.1 requests routed to 1.2, billed at 1.2 rates | Your 1.1 calls start returning 1.2 output |
The Veo row covers veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview, all shut down on 22 October 2026 with Omni 1.2 Flash as the replacement.
The price change in numbers
Omni Flash is billed by tokens, not by clip. Google converts each second of generated video to a fixed number of output tokens and charges the per-million rate. The input rate and the text output rate do not change between versions. All figures below are from Google's pricing and model pages on 17 November 2026; the per-second and per-clip figures are our arithmetic on Google's token rates.
| Item | Omni 1.1 Flash | Omni 1.2 Flash |
|---|---|---|
| Input (text, image, video) per 1M tokens | $1.50 | $1.50 |
| Text output per 1M tokens | $9.00 | $9.00 |
| Video output per 1M tokens | $17.50 | $12.00 |
| Tokens per second of 720p video | 5,792 | 5,792 |
| Effective price per second at 720p | about $0.101 | about $0.070 |
| 10-second 720p clip | about $1.01 | about $0.70 |
| Longest single clip | 10 seconds | 15 seconds |
| Free tier | Not available | Not available |
The working: 5,792 tokens multiplied by $12.00 per million is $0.0695 a second, and by $17.50 per million it is $0.1014 a second. The ratio 12.00 divided by 17.50 is 0.686, so the cut is 31.4%, which Google rounds to 30%.
Resolution changes the bill more than the version does
Google publishes token rates for four output resolutions on 1.2 Flash, and the guide marks 1080p and 4K as upscaled outputs.
| Resolution on 1.2 Flash | Tokens per second | Price per second | 10-second clip |
|---|---|---|---|
| 360p | 1,931 | about $0.023 | about $0.23 |
| 720p (default) | 5,792 | about $0.070 | about $0.70 |
| 1080p, upscaled | 8,688 | about $0.104 | about $1.04 |
| 4K, upscaled | 17,376 | about $0.209 | about $2.09 |
These follow from the token counts on Google's pricing page at $12.00 per million. A 4K clip costs nine times a 360p clip of the same length, because 17,376 divided by 1,931 is 9.0. Google's launch post suggests drafting at 360p or 720p, fixing details with a follow-up edit rather than a fresh generation, and upscaling only the take you keep. Google does not publish a tokens-per-second figure for 1.1 at resolutions other than 720p, so we have not compared those.
A worked monthly bill
Take a team that generates 2,000 clips a month, each 10 seconds at 720p, which is 20,000 seconds of video. This is an illustration, not a benchmark.
- On 1.1 Flash: 20,000 seconds multiplied by $0.10136 is $2,027.20.
- On 1.2 Flash: 20,000 seconds multiplied by $0.069504 is $1,390.08.
- Saving: $637.12 a month on video output alone.
Input tokens are extra on both. Prompts are small, but reference images and uploaded videos for editing count as input at $1.50 per million tokens. Rejected takes cost the same as kept ones, so the number of attempts per usable clip matters more than the version: at 3 attempts per kept clip the 1.2 bill above becomes $4,170.24. Our earlier analysis of what a 60-second brand video actually bills on Omni Flash works through take counts in more detail.
Five things that change on 3 December besides the price
Google says the routing needs no code change, and the request format does not change. The model behind the ID does, and that can surprise a production pipeline.
1. Clips can be longer, so set the duration
On 1.2 you can set duration in response_format to any whole second from 3 to 15. When you leave it out, the model defaults to 10 seconds, and when you set it, the value overrides any length written in the prompt. Code written for 1.1, which topped out at 10 seconds, may not set a duration at all. Set it explicitly so a prompt that mentions a longer scene does not produce a longer, more expensive clip.
2. Audio appears in the output
Google's model page lists the 1.2 output as video with native audio, plus text, and lists the 1.1 output as video only. The developer guide says text-to-video generation produces a video with audio, and suggests negative prompt lines such as no dialogue or no extra sound effects to remove what you do not want. If your pipeline adds its own music or voice-over, check that it strips or mutes the generated track.
3. Some inputs and edits have new limits
1.2 accepts input video of up to 15 seconds for editing and uses the last 10 seconds of a clip when extending it, by 3 to 10 seconds per turn up to 30 seconds in total. Clips of 11 to 15 seconds accept only 1 reference video of up to 3 seconds, while 3 to 10 second clips accept up to 3 reference videos. First and last frame interpolation is limited to 3 to 10 seconds, and uploading audio inputs is not supported in the current API. Editing or extending uploaded videos is not available to users in the European Economic Area, Switzerland and the UK, although editing videos the model itself generated is.
4. Rate limits are per model, and Tier 1 is tight
Google lists these limits for 1.2 Flash: Tier 1 gets 8 requests a minute and 40 a day, Tier 2 gets 16 and 200, and Tier 3 gets 40 and 2,000. The free tier is not supported. At 40 requests a day, a Tier 1 project tops out near 1,200 clips in a 30-day month, and Google lists a $250 billing cap for Tier 1, so the 2,000-clip example above needs Tier 2. Google also says the listed limits are not guaranteed.
5. Outputs are watermarked and filtered
Google says every Omni output carries an invisible SynthID watermark that software can detect, and that safety filters apply to both prompts and generated video, varying by region. English is the only fully evaluated prompt language. A routing switch is a good moment to recheck your block rate on real prompts, since a blocked request still costs you a retry and time.
Gemini API versus Google Cloud
The 3 December routing is documented for the Gemini API, on Google's AI Studio developer pages. On Google Cloud, the Gemini Enterprise Agent Platform release notes record 1.2 Flash as generally available on 5 November 2026 through the Interactions API, with standard pay-as-you-go and Provisioned Throughput at 685 tokens per second per GSU. The same release notes list Omni 1.1 Flash on that platform as a preview under the ID gemini-omni-1.1-flash-preview, and we did not find a 3 December routing note for it there. If you run Omni through Google Cloud rather than an AI Studio key, check that platform's own model lifecycle page and your billing account before assuming the same date.
A test plan for the next two weeks
Today is 17 November, so there are 16 days to 3 December. Use them to see 1.2 output before your users do.
- Find every caller. Search code and configuration for
gemini-omni-1.1-flashand for any older preview ID. Note which calls setduration,resolutionandaspect_ratio.
- Run a side-by-side batch. Send 20 to 50 real prompts to
gemini-omni-1.2-flashwith the same settings. Compare clip length, audio, subject consistency and block rate against your 1.1 results.
- Pin the settings. Set
durationandresolutionexplicitly on every call. Add a prompt line on audio if you do not want generated sound.
- Switch the ID yourself. Moving to
gemini-omni-1.2-flashbefore 3 December gives you a controlled change and a cheaper bill now, rather than an unannounced switch later.
- Watch quota and spend. Check your tier against the daily request cap, and set a budget alert based on seconds generated multiplied by the per-second rate for your resolution.
- Keep a fallback. Veo 3.1 Preview is already gone from the Gemini API, and Google points Veo users to Omni 1.2 Flash or to the GA Veo models on its Gemini Enterprise Agent Platform, so a fallback for critical video jobs means another platform or another vendor. Our note on the Sora 2 and Videos API shutdown shows what happens when a video endpoint closes with no replacement.
A gateway that logs model, resolution and seconds per request makes the rest of this easy; our AI gateway and model-routing service describes how we set one up.
Data use on the paid tier
Omni Flash runs only on the paid tier of the Gemini API. Google's additional terms say that on paid services it does not use your prompts, files or responses to improve its products, and that it processes them under its data processing addendum, while logging prompts and responses for a limited period to detect policy violations. Unpaid use is different: Google may use that content to improve its products, and human reviewers may read it. The terms treat AI Studio as a paid service only when the account has access to a project with an active billing account. Keep client footage and unreleased product images on a paid project.
For teams in India and the UK
At the ECB euro reference rates of 16 November 2026, crossed through the euro and rounded to two decimals, one US dollar buys ₹93.93 or £0.76. On those rates a second of 720p video costs about ₹6.53 or £0.05 on 1.2 Flash, against about ₹9.52 or £0.08 on 1.1. The worked monthly bill drops from about ₹1,90,415 to about ₹1,30,570, or from about £1,541 to about £1,056. Google bills the Gemini API in US dollars, so a weaker rupee raises the rupee cost even when the dollar price falls. If your videos are published in India, check the labelling rules for synthetic media, which do not depend on the model version; our guide to making AI videos for free summarises labelling alongside the free tools. UK teams should note the restriction on editing uploaded videos described above.
How eCorpIT can help
eCorpIT is a Gurugram-based technology organisation, founded in 2021, assessed at CMMI Level 5 and MSME certified, with senior-led engineering teams working across AWS, Microsoft and Google platforms. Our artificial intelligence and machine learning services team can test your Omni Flash prompts on 1.2, pin the settings that control cost and set up spend tracking before 3 December. Talk to us at /contact-us/.
Last updated: 17 November 2026.