AI video cost per minute on Sume: eight models by per-second rate
One minute of generated footage costs $0.75 to $34.67 on Sume depending on the model. Per-second rates and minute totals for eight models.

A minute of generated video costs between $0.75 and $34.67 on Sume, depending on the model and resolution. The numbers below are per-second rates multiplied by 60; real jobs are capped per clip by each model's limits, so a minute means several clips.
Rates and minute totals
Sume bills the provider list price times 1.25, and the docs say that applies to every video model. Rates come from the repo's pricing estimators and are rounded to four decimals where needed.
| Model | Resolution | Per second | Per 60 seconds |
|---|---|---|---|
| Grok Imagine Video | 720p | $0.0125 | $0.75 |
| Wan 3.0 | 480p | $0.0625 | $3.75 |
| MiniMax H3 Max | 768p | $0.10 | $6.00 |
| Wan 3.0 | 720p | $0.125 | $7.50 |
| Kling v3 Pro, audio off | default | $0.14 | $8.40 |
| Seedance 2 | 720p | $0.378 | $22.68 |
| Seedance 2.5 | 720p | $0.578 | $34.67 |
| Wan 3.0 | 1080p | $0.25 | $15.00 |
Reading the table
Resolution moves cost more than most people expect. Wan 3.0 is exactly $0.0625, $0.125 and $0.25 a second at 480p, 720p and 1080p, so each step doubles the price. MiniMax H3 Max is $0.0625, $0.10 and $0.20 at 480p, 768p and 1080p. Seedance models are billed by video tokens, so their prices climb faster with resolution.
What a minute is not
- It is not one request. The docs list 4 to 30 seconds per Seedance 2.5 clip and at most 15 seconds for most other models.
- It does not include editing. Stitching clips is a timeline render at $0.10 per started output minute.
- It does not include retakes. Plan on generating more than you ship.
A cost-per-minute habit worth keeping
When two models look close in quality, convert both to cost per finished minute and include your retake rate. A model at $0.125 a second with a one-in-two keep rate costs $15.00 per kept minute. A model at $0.0625 a second with a one-in-four keep rate costs $15.00 as well. The cheaper rate is only cheaper if it does not need more takes.
That arithmetic is yours to measure; this post does not claim any model's keep rate. It only gives you the dollars-per-second column to multiply.
For a live figure, call the models endpoint the video docs describe. These numbers are a planning grid as of 2026-10-04.
Sources
Related posts
More in Pricing
- AI video price per frame at 24 fps: Wan, Omni, H3 Max, Veo Lite
A 24 fps second is 24 frames. Price per frame: Veo 3.1 Lite $0.0021, Sume Wan 3.0 480p $0.0026, H3 Max 768p $0.0042, Wan and Omni 720p $0.0052.
- AI video price per minute table, next to Sume's per-model SKUs
A Hedra table from Artificial Analysis lists AI video at $6.00 to $20.16 a minute. How to read it, and where Sume prints its own per-model SKUs.
- Audio and video API dates to calendar: Oct 12, Nov 11, end of 2026
ElevenLabs v4's 72% discount ends Oct 12, Seedance 1.5 Pro retires Nov 11, Gemini 3.8 Flash TTS pricing runs to year end. A dated table and a countdown script.
- Batch APIs discount 50%: what a Sume bulk run changes instead
OpenAI, Anthropic and Gemini each document a 50% batch discount. Sume docs describe no batch price: a bulk queue controls throughput, not the rate.
Written by Sume