Gemini Omni Flash price per second: 5,792 tokens at $17.50 per million

Gemini Omni Flash on Google's Gemini API bills video output at $17.50 per million tokens, and one second of 720p video is 5,792 tokens, so a second costs 5,792 x 17.50 / 1,000,000 = $0.1014. Google's page rounds that to about $0.10 per second; on Sume the same model is billed at its $0.10 list rate times 1.25, which is $0.125 per second at 720p.

The Google numbers

The pricing page lists input at $0.75 per million tokens through December 31, 2026 and $1.50 per million from January 1, 2027, and output at $9.00 per million for text and $17.50 per million for video. It states the video billing rate as 5,792 tokens per second of 720p video. Input tokens add a little on top, for example a 300-token prompt adds about $0.0002 at the $0.75 rate.

Check the arithmetic in code

Keep the conversion in one function so a price change on Google's page is a one-line edit. The Sume column uses the documented rule: list price times 1.25, rounded up to cents.

import math

TOKENS_PER_SECOND_720P = 5792
GOOGLE_VIDEO_OUT_PER_M = 17.50
SUME_LIST_720P = 0.10   # per second, Sume Video Router catalog row
SUME_MULTIPLIER = 1.25

for seconds in (3, 8, 10):
    tokens = TOKENS_PER_SECOND_720P * seconds
    google = tokens * GOOGLE_VIDEO_OUT_PER_M / 1_000_000
    sume = math.ceil(round(seconds * SUME_LIST_720P * SUME_MULTIPLIER * 100, 6)) / 100
    print(f"{seconds:>2}s  {tokens:>6} tokens  google ${google:.4f}  sume ${sume:.2f}")

Why the Sume figure is higher

Sume bills the provider list price times 1.25 on each catalog row, and the list price comes from the provider, which for this row is fal (list $0.03, $0.10, $0.15 and $0.30 per second at 360p, 720p, 1080p and 4K, read on 2026-08-28 in the repo docs). The extra 25 percent buys one API for the whole video catalog, async jobs with Idempotency-Key, and signed webhooks. If you only ever call Omni and already hold a Google billing account, the direct token rate is lower.

Using the table for a budget

The comparison also shows why list price alone is a poor guide. A 1,000-clip month of 8-second 720p clips is $811 at Google's token math and $1,000 at Sume's billed rate, a gap of about $189 that you trade for one API, async jobs and signed webhooks.

• Set a per-clip ceiling in code, using the billed column, and refuse a job above it.
• Log usage.cost from every completed job and compare it with your estimate once a week.
• Re-read Google's pricing page monthly, because the token rate and the tokens per second can change.
• Do not mix standard and any batch or flex rates in one estimate.

What you see on a real job

A completed poll response from GET /v1/videos/{id} carries usage.cost, which is the Sume billable amount, so you can reconcile the table against your own jobs. Sume reserves the estimate at submit, which is why a balance below the reserve returns 402 insufficient_credits rather than a failed job later. See Errors and credits for the codes.