Omni references on Sume: billed per output second; Google adds input

Google lists $1.50 per 1M input and $17.50 per 1M output tokens for Omni. Sume bills the provider list x 1.25 per output second. Reference limits compared.

4 min readSume
All posts

On Google's price list, Gemini Omni Flash charges $1.50 per million input tokens (text, image, video and audio) and $17.50 per million output tokens for video. On Sume, the docs say the billed quantity for gemini-omni-flash-1.1 is each output second at the provider list times 1.25, so the number of reference images you attach is not a separate line. The limits are the same family: up to 10 reference images and up to 3 reference videos of 3 seconds each.

The two price structures

Google bills in tokens, with separate input and output rates. Its pricing page says output is billed at 5,792 tokens per second of 720p video, and gives no per-image or per-second input price for references. Sume's docs, for the Omni row, state a single basis: the provider list times 1.25 for each output second, as a function of resolution.

How references are priced for Omni Flash, pages read 2026-10-08
ItemGoogle (Gemini API)Sume
Output video$17.50 per 1M tokens$0.0375 to $0.375 per second by resolution
Input (text, image, video, audio)$1.50 per 1M tokensNo separate input line in the docs
Reference imagesImage references supportedUp to 10, tagged <IMAGE_REF_0>
Reference videosMax 3 clips, up to 3 s eachUp to 3 clips, each up to 3 s, tagged <VIDEO_REF_0>
Billing unitTokensOutput second

What it means for a prompt with many references

On Sume a job with 1 reference image and a job with 10 reference images, both 8 seconds at 720p, are priced as 8 x $0.125 = $1.00 in the docs' model. On Google the references add input tokens, and the amount depends on the size of each file. Google's pages do not give a fixed number for an image, so any Google comparison here would be a guess, and this post does not make one.

The practical advice is the same on both: add references because they improve the clip, not because they are free. The cost of a bad clip is the output seconds you re-render.

Limits to keep in mind

Reference videos are limited to 3 clips of 3 seconds, so trim a longer source first with Sume's video trim. Refer to media in the prompt as <IMAGE_REF_0> and <VIDEO_REF_0>, 0-based in list order. You cannot send video_url (the edit source) together with image_url, end_image_url or the reference fields. There is no reference_audio_urls on this model.

Read usage.cost on the first completed job with references to confirm that the amount matches output seconds times the rate. If it does not, the docs are the place to start, and the support route is the place to finish.

A way to test it yourself

Run two 5-second jobs at 720p with the same prompt, one with no reference and one with three reference images. Each should read 5 x $0.125 = $0.625, billed as $0.63 after cent rounding, if the docs' billing basis holds. Keep the two usage.cost values in your notes. If they differ, tell Sume support and include both job ids.

A test like this is cheap, about $1.26 for the pair, and it settles the question for your own prompts.

Sources

Related posts

More in Pricing

All Pricing posts

Written by Sume