Gemini Omni 1.2 Flash API: Model ID, Cost per Clip, 1.1 Routing

The Gemini API model ID for Gemini Omni 1.2 Flash is gemini-omni-1.2-flash. Google made it generally available on November 5, 2026, for the same job as before, generating video with native audio and editing it in follow-up turns, but with clips up to 15 seconds instead of 10 and video output billed at $12.00 per million tokens instead of $17.50. At 720p that is about $0.07 per second of video, according to the Gemini API pricing page. There is no free tier: both Omni versions need a Gemini API key on a paid billing tier.

If your code calls gemini-omni-1.1-flash, it keeps working, but not on 1.1 for much longer. The Gemini API release notes say that from December 3, 2026, every request to that ID is automatically routed to gemini-omni-1.2-flash and billed at 1.2 rates, with no code change and no error. Your requests will then be answered by a different model, so change the model string to gemini-omni-1.2-flash now and rerun your own prompts while 1.1 is still available to compare against.

Two kinds of users should look past the Gemini API. If you need Provisioned Throughput or Cloud controls such as data residency and CMEK, Google Cloud's Gemini Enterprise Agent Platform serves the same model ID with those options. If nobody on the team writes code, Flow and Google Vids run Omni 1.2 without an API key, and the Gemini app is rolling it out to Google AI Ultra subscribers. Prices, limits and IDs below are as of November 6, 2026.

What changed in Gemini Omni 1.2 Flash vs 1.1

For code, the 1.2 changes come down to longer clips, a lower price and a few new caps. Output can now be any whole number of seconds from 3 to 15, set with duration in response_format, and you can edit input videos up to 15 seconds long. Extension is now capped at 30 seconds of total length. Clips longer than 10 seconds accept only one reference video, and every request can carry up to 6 reference images. The Gemini Omni Flash model page lists both IDs as Stable, and Google's release notes say paid-tier rate limits doubled with 1.2.

The 1.2 values come from the Gemini API Omni guide and the model page, and the 1.1 values from the 1.1 model card and Google's 1.1 launch material. Only one row can break a pipeline that works on 1.1 today: an extension chain that relied on the 40-second total stops at 30 seconds on 1.2. The reference-video and interpolation caps start to matter once you lengthen clips past 10 seconds. The price cut is 31.4% (12.00 ÷ 17.50), which Google rounds to 30%.

The chart below was drawn for the 1.1 release in August 2026. Its 40-second extension total and 10-second upload limit are the 1.1 rules; use the table above for 1.2.

What happens to gemini-omni-1.1-flash on December 3, 2026

Nothing breaks on December 3. Requests that name gemini-omni-1.1-flash still succeed, but Google serves them with gemini-omni-1.2-flash and charges 1.2 rates. The Omni guide describes the weeks until then as a 4-week transition window. After that date, the Gemini API documents no way to get 1.1 output, and Google doesn't say whether a routed response will report 1.1 or 1.2 in its model field, so don't build a check on it.

That makes the choice simple for most teams: switch now, on a date you control, rather than have the model change underneath you on December 3. Letting the routing happen is reasonable only when you generate occasional clips and judge them by eye, because you will get the lower price either way. If clips go to customers, feed an editing pipeline or depend on a fixed length, test these before December 3:

1. Rerun 5–10 of your real prompts on both IDs at the same resolution and compare character consistency, motion and audio side by side.
2. Set duration explicitly wherever a clip must have a fixed length. The default is still 10 seconds, and an explicit value overrides any length written in the prompt.
3. Find extension chains that go past 30 seconds of total length and shorten or split them.
4. If you lengthen clips past 10 seconds, pass at most one reference video and keep first-and-last-frame requests at 10 seconds or less, since two frames with a longer duration return an invalid argument error.
5. Update cost alerts and dashboards that key on the model string, since spend per clip drops about 31% at every resolution.

Make a first Gemini API request

A first Omni 1.2 request needs a paid-tier API key, Google's Gen AI SDK and one interactions.create call. Omni uses the Interactions API, so code written around generateContent calls for text or image models won't carry over unchanged.

1. Create an API key in Google AI Studio for a project with billing turned on. The free tier doesn't support Omni.
2. Install the SDK with pip install -U google-genai and set the key as the GEMINI_API_KEY environment variable. If GOOGLE_API_KEY is also set, the SDK uses that one instead, so remove it or make both match.
3. Run the request below, keep the returned interaction.id, and pass it as previous_interaction_id with a new input when you want to edit the clip.

import base64
from google import genai

client = genai.Client()  # reads GEMINI_API_KEY

interaction = client.interactions.create(
    model="gemini-omni-1.2-flash",
    input=(
        "A continuous, unbroken shot of a paper model city waking at dawn. "
        "Warm window lights switch on one by one. No dialogue."
    ),
    response_format={
        "type": "video",
        "aspect_ratio": "16:9",   # or "9:16"
        "resolution": "720p",     # 360p, 720p, 1080p or 4k
        "duration": "15s",        # "3s" to "15s"; default "10s"
    },
)

with open("city.mp4", "wb") as f:
    f.write(base64.b64decode(interaction.output_video.data))

print(interaction.id)  # use as previous_interaction_id for the next edit

interaction.output_video exists only in the SDKs. If you call REST directly with POST https://generativelanguage.googleapis.com/v1beta/interactions?key=YOUR_API_KEY, the base64 video sits in the steps array, inside the step of type model_output, in a content item of type video. For clips over 4 MB, which Google ties to resolutions above 720p, add "delivery": "uri" to response_format, poll the returned file until its state is ACTIVE, then download it. Take the URI from the creation response, because a later GET on the interaction returns inline base64 instead. Google suggests store=false for faster synchronous calls, but a clip generated that way can't be edited in a later turn.

Start at 360p or 720p to judge composition, and move to 1080p or 4K only for a take worth keeping. Google calls both higher resolutions upscaled output, so 4K here is not native 4K detail.

Gemini Omni Flash API key: no free tier, and what 400 and 403 mean

Gemini Omni Flash is not free on the API. Both the pricing page and the Gemini API rate limits page list the free tier as not available for Omni, so the key must belong to a project with billing. The free way to try Omni 1.2 that Google names is Google Vids, which makes 15-second clips at no additional cost with a Google account or a Workspace subscription. For how the Gemini API free tier works on other models, see Gemini API Free Tier 2026: Limits, Free Models, and Free API Keys.

On November 6, 2026, the request above was sent to the live endpoint from Python 3.12.1 with google-genai 3.0.0, the newest release on PyPI that day, using a deliberately invalid key. The SDK accepted the documented call shapes without a local error: a 15-second 720p generation, an edit turn with previous_interaction_id, and the old gemini-omni-1.1-flash ID. Google's server answered each with 400 INVALID_ARGUMENT and "API key not valid. Please pass a valid API key." (reason API_KEY_INVALID), which means the key itself is wrong. The same REST POST with no key at all returned 403 PERMISSION_DENIED and "Permission denied on the service." No video was generated and nothing was billed, so this test confirms the request format and the endpoint, not generation time, output quality or the actual bill.

How much a Gemini Omni 1.2 clip costs: about $0.07 per second at 720p

Omni bills the video by output tokens, and the number of tokens depends only on resolution and length. Multiply the tokens per second from Google's pricing footnote by $12.00 per million on 1.2, or $17.50 on 1.1, and by the clip length. A 10-second 720p clip on 1.2 is 5,792 × 10 × $12.00 ÷ 1,000,000 = $0.695. The table applies that formula to every resolution; it was computed on November 6, 2026, and matches Google's own rounded figures of about $0.07 per second and about $1.04 per 15-second clip at 720p.

For a budget, what matters is how many clips you get. Dividing $10 by the output cost per clip and rounding down gives these counts:

The workflow saves more than the price cut. Google's launch post recommends drafting at 360p or 720p and moving to 1080p only for the take you keep. Four 10-second drafts at 360p plus one final 10-second clip generated at 1080p cost $0.927 + $1.043 = $1.97, while the same five attempts all at 1080p cost $5.21, so drafting first spends 62% less. The final here is a fresh 1080p generation, because Google doesn't publish whether upscaling an existing clip is billed the same way.

All of these are list prices for output only. Input tokens cost $1.50 per million on top, for reference images and uploaded video, and the tables leave out retries, blocked generations and tax. Google's pricing documentation also doesn't say whether an extension turn bills only the new seconds or the whole returned video. If only the new seconds count, a 30-second 720p piece built as 15 + 10 + 5 seconds would cost about $2.09 in output, but treat that as an estimate until your billing report confirms it.

If you are comparing Omni with Veo 3.1, which you now buy through Google Cloud or Flow credits rather than the Gemini API, see Veo 3 Price 2026: Cloud API Pricing, Flow Credits, Cost per Video.

Gemini Omni 1.2 rate limits: 8 RPM and 40 requests a day on Tier 1

On Tier 1, the daily request cap limits a batch before the budget does. The Gemini API rate limits page gives gemini-omni-1.2-flash 8 requests per minute and 40 per day on Tier 1, and every edit or extension turn is its own interactions.create call, so it uses a request too. Google adds that the listed limits are not guaranteed, and AI Studio shows the limits active on your project.

The last column is requests per day × $0.69504, the cost of one 10-second 720p clip. If each finished clip takes four drafts and a final, Tier 1 yields 8 finished clips a day and Tier 2 yields 40, so a 50-clip campaign takes 7 days on Tier 1 and 2 days on Tier 2. Google's release notes call the 1.2 numbers double the old paid-tier limits, but the page no longer lists the 1.1 values.

Check the limits before uploading footage

Where you are and what you upload matter more than any other limit. In the European Economic Area, Switzerland and the United Kingdom, the Omni guide says editing or extending an uploaded video is not available, although editing and extending clips the model generated still works there. The same three regions can't upload or edit images that contain minors, and uploading or editing images of certain recognizable people isn't supported anywhere.

The other limits that change a plan are about inputs. Reference videos work only for new generations, not edits or extensions, and any audio in them is ignored; reference images work for generation, editing and prompt-based extension. Extension only appends to the end of a clip and uses its last 10 seconds as context, so mention any character who appeared earlier again in the prompt. You can't add new dialogue when extending an uploaded video in which someone is talking, but multi-turn extension of a generated clip can add speech. The Gemini API also rejects audio inputs and audio references, can't edit voices, and has no system instructions, temperature, top_p, stop sequences or negative prompts, so write "Do not…" into the prompt itself. It doesn't reason across several videos at once, doesn't take YouTube URLs as a source, and doesn't offer Provisioned Throughput.

English is the only fully supported prompt language; Google hasn't evaluated the others, so they may work with varying results. Every output carries an invisible SynthID watermark, and safety filters, which differ by region, apply to both prompts and videos.

Which Omni model ID to use in each Google product

On the Gemini API and in Google AI Studio, use gemini-omni-1.2-flash. Google Cloud's Gemini Enterprise Agent Platform now uses the same ID for 1.2, unlike 1.1, which Cloud offered only as gemini-omni-1.1-flash-preview. The older gemini-omni-flash-preview is shut down on the Gemini API. Flow, Google Vids and the Gemini app have no model-ID field at all; you get Omni 1.2 through the product and your plan.

The API is the wrong choice in two cases. Pick the Agent Platform when you need reserved capacity or enterprise controls, because the Gemini API guide lists Provisioned Throughput as unsupported, while the Gemini Enterprise Agent Platform model page for Omni 1.2 Flash lists it alongside standard pay-as-you-go. On Cloud, too, Omni runs only through the Interactions API; the GenerateContent and Chat Completions APIs are not supported, and batch inference isn't either. Pick Flow, Vids or the Gemini app when the work is interactive and nobody needs to call a model from code. Google hasn't published Flow credit costs for 1.2 yet, and how many clips each app allows per day is covered in Google Veo Limit Per Day: 5 Clips in Flow, No Count in Gemini.

The overview below was made for the 1.1 release, so its model IDs and Cloud preview status are the 1.1 ones; the product split it shows still applies.

Where Google's Omni 1.2 pages disagree

Google's launch post and Cloud model page describe a few features differently from the Gemini API guide, mostly because each page describes a different product: the Omni 1.2 launch post on the Google blog covers every Google product at once, and the Cloud page covers only the Agent Platform. When you call the Gemini API, go by the Gemini API guide.