Gemini Omni Pricing: 17 Test Clips, Costs and Catches

Google shipped Gemini Omni 1.2 Flash on 5 November 2026 and cut the video output rate from $17.50 to $12.00 per million tokens. On Google’s own arithmetic, that takes 720p video from about $0.10 a second to about $0.07. Gemini Omni pricing is quoted in tokens, though, and Omni also bills for thinking and for re-reading video when you edit, so the rate card doesn’t quite tell you what a clip costs.

So we generated 17 real clips on our own Gemini API key, the day after launch, at every resolution and at lengths from 3 to 15 seconds, and added up what each one consumed. We wanted the figures a WordPress store owner needs before pointing this at a product catalogue: the cost per clip, how long you wait, how big the file is, and whether the product in the video is your product.

The short answer: a 5-second 720p clip costs $0.35 on 1.2 Flash against $0.51 on 1.1 Flash, and a full 15-second clip costs $1.05. Billed video tokens matched Google’s published rates exactly on every clip. The cost that isn’t on the rate card is retakes: our text-only prompt produced a different mug every time, so plan on starting from a product photo, and budget for more than one take.

Gemini Omni pricing: 1.2 Flash vs 1.1 Flash

These are the paid-tier rates on Google’s Gemini API pricing page, read on 6 November 2026, per million tokens. Neither model has a free tier, and only Standard pricing is listed, with no batch tab.

Google’s changelog calls this “30% lower”; $17.50 to $12.00 is 31.4%. On 3 December 2026, requests to gemini-omni-1.1-flash start routing to 1.2 Flash automatically and bill at the 1.2 rates. If you stay on the old ID until then, you pay the old price for a model Google is already moving you off. Changing the ID today saves about 31% on video output.

Video output is billed by resolution and length, not by what’s in the frame. Google converts seconds to tokens at a fixed rate:

Clip length can be any whole second from 3 to 15, and the default is 10 seconds. That default is worth noticing: leave the duration out and you buy a $0.70 clip even if you only wanted a short loop.

How we tested it

All clips came from a PHP script on our WordPress server, calling Google’s Interactions API directly with the same key our site uses. We used two product-video prompts, one for a blue ceramic mug and one for a tan leather backpack, both asking for a single unbroken scene with no people and no on-screen text.

• Resolution: 5 seconds on 1.2 Flash at 360p and 1080p (both prompts) and 4K (mug only), plus 720p with both prompts and one repeat of the mug.
• Old vs new: the same two prompts at 720p on 1.1 Flash.
• Length: the mug at 720p for 3, 10 and 15 seconds, plus a 3-second 360p probe.
• Editing: one colour change on an existing clip.
• Image to video: one uploaded product still, run twice.

That makes 17 clips, and about $7.57 at list rates. Google returns token counts rather than a price per request, so costs below are the reported tokens multiplied by the published rates. We priced thinking tokens at the $9.00 text rate, following the pricing page’s “including thinking tokens” line, and we didn’t reconcile individual clips against an invoice. Most setups ran once or twice, so treat the timings as a sample from one morning.

What each clip cost

Three things stand out in those numbers.

The token maths is exact. Every clip reported precisely Google’s rate: 28,960 video tokens for 5 seconds at 720p, 86,880 for 15 seconds, 9,655 for 5 seconds at 360p. Files ran about 0.03 seconds longer than requested, at 24 frames per second, with H.264 video and AAC audio. Nothing varied with the prompt, so video output is one of the easier AI costs to forecast.

Thinking is a small extra. Each clip also used 234 to 1,146 thinking tokens, about 600 on average, which adds half a cent or so. At 720p and 5 seconds that’s 1% to 2% on top of the video output. At 360p, where the video is cheap, it reached 8%. Input barely registers: a text prompt of about 60 tokens costs a hundredth of a cent, and a 1280 × 720 reference image came in at 1,100 tokens, or $0.0017.

1.1 Flash wasn’t faster in exchange for the higher price. On the same two prompts, 1.1 took 23.0 and 37.0 seconds against 25.1 and 30.4 on 1.2. Two clips per setup can’t rank them, but we saw no reason to stay on the old ID.

For scale, a 5-second 720p clip costs about the same as 415 support answers on Claude Haiku 5.5, at the $0.85 per 1,000 we measured in our Kimi K3.1 Pro providers test. Video is priced like a marketing asset, so it belongs in that budget, not in a per-conversation one.

The hidden cost: your product might not be in the video

Prompt-only video is fine for mood and background footage. It’s a different matter for a shop. We asked for one hand-thrown blue mug on an oak table, and this is what came back:

All six fit the description, and no two are the same mug. Three came with a saucer or plate we didn’t ask for, and two of those added a spoon. If a customer can buy that mug, none of these clips show it. Google’s prompt guide says Omni tries to craft a narrative by default, and models fill gaps. That’s useful for creative work and awkward for a catalogue.

Starting from a photo mostly solves it. We took one still of the backpack, uploaded it, and asked for a slow orbit with the backpack kept identical:

The side we supplied held up well in both runs: same colour, straps, buckles and pockets, on the same plinth. The back didn’t, because the model had never seen it. Each run invented its own back panel and straps. So use a photo for anything you sell, and either keep the camera on the side you photographed or supply more angles. Google allows up to 6 reference images. We ran only one, so we can’t say how well six would hold up. Image input added less than a cent, so this is cheap protection.

Here is one of those two runs, as the API returned it, with audio:

Editing costs a little more than a new take

You don’t have to regenerate to fix something. We asked for the same mug in deep red, using previous_interaction_id on a stored clip:

The table, window, light and framing all held, and the colour changed. The turn cost $0.39 against $0.35 for a fresh clip, because Omni re-reads the earlier clip as 28,960 tokens of video input at $1.50 per million. That adds about $0.04. It took 45.8 seconds, roughly 15 seconds longer than a fresh take. When you have a clip you like and want one change, editing is the better buy, because a new take will give you a new mug.

Editing needs store left on, which matters if you’ve copied Google’s speed tip of turning it off. The same setting caught us at higher resolutions: the API rejected our first 1080p requests with “store=true is required when response format has video delivery set to URI”. Google’s docs recommend URI delivery above 4 MB. In practice our 5.3 MB 15-second clip still came back inline without complaint, and a URI file took about 5 seconds more to become ready for download.

What this means for a WordPress site

Don’t generate inside a page request. Nothing here finished fast enough to run while someone waits on a normal PHP request. Even the small clips took 15 seconds, and our 4K clip took 110, past the 100-second point where Cloudflare’s proxy gives up on a request. Run generation as a background job and save the result when it arrives.

Skip 4K for web use. It costs three times the 720p price, comes from upscaling rather than generation, and produced 15.6 MB for 5 seconds, or about 3.1 MB per second. At that rate a 15-second 4K clip would be around 47 MB (our estimate; we didn’t render one), which is over the 32 MB upload limit on our own site. A 720p clip runs at about 0.35 to 0.57 MB per second, which is far easier to host. For a product page, 720p is the sensible default. Move to 1080p only if the clip will be shown large.

Watch the daily limits before a bulk run. Google’s rate limits page lists these for 1.2 Flash, doubled at launch according to the changelog:

Budget by takes, not by clips. This is the sum for a store making one short clip per product, using our measured image-to-video cost of $0.354 for 5 seconds at 720p, and $0.51 for the same length on 1.1 Flash. How many takes you need depends on your products and how fussy you are, so we show one and three:

For most small stores, spend isn’t the limit. Reviewing clips is, along with checking that each one shows what you actually ship. A clip that shows a different product is a returns and trust problem. Every generated video also carries Google’s invisible SynthID watermark, which can be detected programmatically.

Omni does all this with sound. Every clip we rendered had an audio track, and Google’s pricing lists no separate audio charge. The mug clips we measured asked for quiet room tone and came out very quiet, about −39 to −47 dB on average, while the backpack clip’s music averaged about −23 dB. Check the audio before you publish, and mute it if the video will autoplay on a product page.

Using Omni from WordPress with MxChat

Our MxChat Video Generation add-on generates clips from your dashboard and saves them to the media library, and it now uses Omni on the Google route because Google retired the Veo models on the Gemini API this year. One honest note: the copy on our own site calls gemini-omni-1.1-flash, so it pays the $17.50 rate until Google’s automatic switch on 3 December. It drops to the lower rate then, with no change on your side. It also leaves 4K out, which matches what we’d recommend anyway.

Omni is a marketing tool, and it has no bearing on what your chatbot costs to run. For that, see Gemini 3.8 Flash pricing and thinking tokens and Gemini Flash-Lite pricing. Our Gemini TTS pricing test looks at another per-second cost billed through tokens, and our Vertex AI pricing comparison covers buying Google’s models through Google Cloud. The add-on is part of MxChat Pro and runs on your own API keys.

Frequently asked questions

How much does Gemini Omni cost per second?

On 1.2 Flash, about $0.023 at 360p, $0.07 at 720p, $0.104 at 1080p and $0.209 at 4K, for video output. Thinking and input added 1% to 2% at 720p in our test.

How much is a 15-second Gemini Omni video?

We measured $1.05 for 15 seconds at 720p on 1.2 Flash. Google quotes about $1.04, which covers the video output alone.

Is Gemini Omni free?

Not on the API. Google lists both Omni models as “Not available” on the free tier, and lists 1.2 Flash as “Not supported” in the free rate limits.

What happens to Gemini Omni 1.1 Flash?

From 3 December 2026, requests to gemini-omni-1.1-flash route to 1.2 Flash and bill at 1.2 rates. You don’t have to change code, but changing the ID now gets the lower price immediately.

Does it cost more to edit a video than to generate one?

Slightly. Our 5-second 720p edit cost $0.39 against $0.35 for a new clip, because the earlier video counts as input tokens at $1.50 per million.

Can I use Gemini Omni through OpenRouter?

OpenRouter lists both models at $1.50 / $9.00 with Google AI Studio as the only host for 1.2 Flash. Its model record doesn’t show the $12.00 video rate, and we didn’t run video through it, so we can’t say what it bills per clip.

Which resolution should I use on a WordPress site?

720p for most product and landing pages. It’s the highest resolution Google doesn’t describe as upscaled, files stay around 2.4 MB for 5 seconds, and it costs a third of 4K.