AI Video Costs $0.035 to $0.57 a Second. Veo 3.1 Charges 2x the #1 Model and Ranks 25th of 34.
Google’s Veo 3.1 costs $0.40 per second of video, twice the $0.20 of Wan 3.0, and ranks 25th of 34 on Artificial Analysis’s text-to-video arena, where Wan 3.0 ranks first. Across the 32 models there with an API price, a second of video costs $0.035 to $0.57, a 16x spread, and price tells you almost nothing about rank.
Where the numbers come from
Artificial Analysis runs a blind video arena: voters see two clips made from the same text prompt, with sound on, and pick the better one. Its AA-Video-T2V v2.1 board turns those votes into Elo scores, with Kling 3.0 1080p (Pro) fixed at 1000 as the anchor. Next to each model it lists what 1 minute of video costs on the model creator’s own API at the settings it benchmarks, 1080p where the model offers it.
That makes it the closest thing to one price list for AI video in November 2026: one methodology, one resolution target, 34 models. Two of them have no price: Utopai X, #2, has no API, and MAGI-2 Preview is “coming soon.” We divided the per-minute prices by 60 for the per-second column and by 6 for a 10-second clip.
The price table: 18 models, ranked by votes
| Rank | Model | Elo | Per minute | Per second | 10-second clip |
|---|---|---|---|---|---|
| 1 | Wan 3.0 | 1175 | $12.00 | $0.20 | $2.00 |
| 3 | MiniMax H3 (768p) | 1161 | $4.80 | $0.08 | $0.80 |
| 4 | Dreamina Seedance 2.5 | 1157 | $34.12 | $0.569 | $5.69 |
| 5 | Gemini Omni Flash 1.2 | 1152 | $6.26 | $0.104 | $1.04 |
| 6 | FLUX 3 Max | 1152 | $8.70 | $0.145 | $1.45 |
| 7 | HeyGen Video 1 | 1151 | $2.40 | $0.04 | $0.40 |
| 8 | MiniMax H3 Max | 1150 | $9.60 | $0.16 | $1.60 |
| 9 | FLUX 3 | 1138 | $17.40 | $0.29 | $2.90 |
| 10 | SkyReels V4 | 1136 | $21.00 | $0.35 | $3.50 |
| 11 | Dreamina Seedance 2.0 | 1135 | $22.45 | $0.374 | $3.74 |
| 12 | P-Video-2-Pro | 1129 | $5.70 | $0.095 | $0.95 |
| 13 | Gemini Omni Flash 1.1 | 1127 | $9.12 | $0.152 | $1.52 |
| 14 | Pika H3 Flex | 1125 | $2.10 | $0.035 | $0.35 |
| 21 | Kling 3.0 Omni 1080p (Pro) | 1020 | $8.40 | $0.14 | $1.40 |
| 22 | Kling 3.0 1080p (Pro) | 1000 | $10.08 | $0.168 | $1.68 |
| 25 | Veo 3.1 | 963 | $24.00 | $0.40 | $4.00 |
| 26 | Veo 3.1 Fast | 955 | $7.20 | $0.12 | $1.20 |
| 27 | Veo 3.1 Lite | 948 | $4.80 | $0.08 | $0.80 |
Artificial Analysis, AA-Video-T2V v2.1, overall category, opened November 15, 2026. Price per minute is the board’s; per second and per clip are our division. Utopai X (#2, 1172) has no API and is left out.
Price barely predicts rank
The top 14 models sit inside 50 Elo points, 1125 to 1175, and their confidence ranges overlap: the board itself gives ranks 3 to 9 as ranges like “3-8” and “3-9.” Inside that near-tie, prices run from $2.10 to $34.12 a minute.
Dreamina Seedance 2.5 costs 7.1x MiniMax H3 for 4 fewer points. It is the most expensive model on the board at $0.569 a second and ranks 4th; MiniMax H3 at 768p ranks 3rd for $0.08. (MiniMax’s price is for 768p, below the 1080p most rows use.)
HeyGen Video 1 ranks 7th for $0.04 a second. Pika H3 Flex, the cheapest model with a price, ranks 14th for $0.035. A 10-second clip on either costs less than 50 cents.
Veo 3.1 is the expensive outlier at the bottom. Google’s January model costs $24 a minute and ranks 25th, 37 points under the Kling 3.0 anchor. Google’s newer Gemini Omni Flash 1.2, released November 5, ranks 5th at about a quarter of Veo 3.1’s price.
On this board the model voters like best costs $0.20 a second, the 4th best costs $0.57, and the 7th best costs $0.04. Budget by testing 2 or 3 models in the top 14 on your own prompts, not by price tier.
Why your bill won’t match the board
The board prices one setting per model. Your invoice depends on three more things.
Resolution. Alibaba’s own price list for Wan 3.0 (Singapore region) is per second of output:
| Wan 3.0, per second | 480p | 720p | 1080p |
|---|---|---|---|
| wan3.0-video | $0.05 | $0.10 | $0.20 |
| wan3.0-video-prime | $0.068 | $0.14 | $0.28 |
Alibaba Cloud Model Studio pricing. The standard model is marked “limited-time 30% off” with no end date given, and includes a 30-second free quota; Prime has none.
A 720p draft costs half the 1080p final. If you generate drafts to pick a direction, render them small.
Tokens, not seconds. Google bills Gemini Omni 1.2 Flash by output tokens: $12.00 per million video tokens, 5,792 tokens per second of 720p. Google puts that at about $0.07 a second at 720p, $0.023 at 360p, $0.104 at 1080p with upsampling and $0.209 at 4K. Clips run up to 15 seconds, and a 15-second 720p clip comes to about $1.04 by Google’s math. The older Omni 1.1 Flash bills $17.50 per million, about $0.10 a second at 720p, and from December 3, 2026 Google routes every 1.1 request to 1.2 at 1.2’s rates.
Inputs and extras. Alibaba bills input video by its duration on top of output, so editing or extending a 10-second clip costs more than generating one. Failed requests aren’t billed there. Audio is often a separate line: on MegaBrain’s Kling pages, turning sound on adds 50% or more to the per-second price.
Sora left the price list
Sora 2 Pro was the reference for premium AI video: $0.30 a second at 720p, $0.50 at 1024p and $0.70 at 1080p. OpenAI’s model page now says the Sora 2 models and the Videos API “were shut down on September 24, 2026.” If an old cost model still has Sora in it, the $0.70 line is gone; the most expensive model on today’s board is $0.569.
Which model for which job
Read the table by job, not by brand:
High-volume social clips and drafts: HeyGen Video 1 ($0.04 a second, #7) and Pika H3 Flex ($0.035, #14) are the only models in the top 14 under 5 cents a second.
Best ranked with an API: Wan 3.0 (#1, $0.20) at 1080p, or MiniMax H3 (#3, $0.08) if 768p is enough for a phone screen.
Already on Google: Gemini Omni Flash 1.2 ranks 5th at $0.104 a second; Veo 3.1 ranks 25th at $0.40. Even Veo 3.1 Lite, at $0.08, ranks below Omni Flash 1.2.
Already on Kling 3.0: the Omni 1080p (Pro) variant scores 1020 to the standard 1080p (Pro)’s 1000 and costs $8.40 a minute to $10.08.
What real deliverables cost
Our arithmetic at the board’s prices, assuming every generation is a keeper:
| Model | 10-second clip | 30-second ad | Week of Reels, 7 × 10 s |
|---|---|---|---|
| Pika H3 Flex | $0.35 | $1.05 | $2.45 |
| HeyGen Video 1 | $0.40 | $1.20 | $2.80 |
| MiniMax H3 (768p) | $0.80 | $2.40 | $5.60 |
| Gemini Omni Flash 1.2 | $1.04 | $3.13 | $7.30 |
| Kling 3.0 1080p (Pro) | $1.68 | $5.04 | $11.76 |
| Wan 3.0 | $2.00 | $6.00 | $14.00 |
| Veo 3.1 | $4.00 | $12.00 | $28.00 |
| Dreamina Seedance 2.5 | $5.69 | $17.06 | $39.81 |
Per-minute prices from Artificial Analysis, divided by our math. Some models cap clip length below 30 seconds, so a 30-second ad may be several clips.
The keeper assumption is the one that breaks budgets. If you keep 1 clip in 3, multiply every cell by 3. A $0.40 clip you reroll 3 times costs $1.20; a 10-second Seedance 2.5 clip costs $17.06. On cheap models you can afford to explore, and on expensive ones you want the prompt locked before you render.
What this means if you build with AI media
Store the price per second next to the model ID. Prices on this board span 16x, and models change rank every month: 6 of the 34 were added in the last 30 days. Put model and price in config so a swap is one line and your cost estimate updates with it.
Draft cheap, render once. Explore with a $0.04 to $0.10 model or at 480p/720p, then render the chosen prompt on the model and resolution you ship.
Turn audio off unless you need it. If you add a voiceover or music in the edit, a native-audio line on the bill is money for sound you’ll mute.
Show users the cost before they click. If your app lets people pick a length, compute seconds × price and display it. A 15-second clip at $0.57 is $8.53, and users should see that before your balance does.
For a working example, our product photo to 9:16 video tutorial builds a 2-call pipeline in Python with the cost of each run, and the media API docs show the queue endpoints. If you’d rather have an agent write it, the Claude Code setup guide shows how to run Claude Code on the same key.
Try it on MegaBrain
Kling 3.0 Pro, from the table above, is available on MegaBrain alongside two Kling models the arena doesn’t list. Kling 3.0 Standard costs $0.0966 per second of video, $0.1449 with sound, so a 5-second clip costs $0.483. Kling 2.6 Pro costs $0.0805 per second, $0.161 with sound. You’re charged when the job is queued, through one API key and one pay-as-you-go balance.
Sources
Pages opened November 15, 2026. Every number above comes from these pages or MegaBrain’s model pages, or from our arithmetic on them where we say so.
- Artificial Analysis — Text to Video leaderboard, AA-Video-T2V v2.1 (Elo, 95% intervals, samples, release month, API price per minute at benchmark settings)
- Alibaba Cloud Model Studio — model pricing (wan3.0-video and wan3.0-video-prime per second by resolution, free quota, discount)
- Google AI Studio — Gemini API pricing (Gemini Omni 1.2 Flash and 1.1 Flash video token rates, December 3 routing)
- Google Cloud — Gemini Omni 1.2 Flash model page (release date, inputs and outputs)
- OpenAI — Sora 2 Pro model page (per-second prices and the September 24, 2026 shutdown)
Full disclosure: MegaBrain sells pay-as-you-go API access to image, video and text models, including Kling 3.0 and Kling 2.6. MegaBrain is not affiliated with Alibaba, ByteDance, Google, MiniMax, HeyGen, Pika, Kuaishou, OpenAI, Black Forest Labs, the other labs whose models appear in the tables, or Artificial Analysis.
MegaBrain for Mac · an AI coworker
Give MegaBrain the task, not the steps.
It works on your files, writes documents, spreadsheets and decks, connects to your team’s tools and asks before anything risky. Free to start.
Newsletter
Stay in the loop
Get our latest AI research and guides. No spam, unsubscribe anytime.