Most AI video models list between $0.05 and $1.164 per second of output, depending on the model, the provider and whether audio is included. The cheapest listing in my data is Runway's Gen-4 Turbo at $0.05 per second, and the most expensive is Seedance 2.5 on fal.ai at $1.164 per second. The same model can cost different amounts at different providers, and every retry is another full render on your bill.
Providers bill video by the second, so I show list prices per second. To get a per-minute figure, multiply by 60. Prices are official or provider list prices as of 2026-10-11, in USD. I did not measure video generation myself. My measured runs cover text tasks, so these are list prices, not observed bills.
List prices per second
One representative listing per model and provider, sorted by price.
| Model | Provider | Per second |
|---|---|---|
| Gen-4 Turbo | Runway | $0.05 |
| MiniMax H3 (768p) | fal.ai | $0.06 |
| LTX-2.3 Pro (1080p) | LTX | $0.08 |
| MiniMax H3 (768P) | MiniMax | $0.08 |
| Veo 3.1 Lite | $0.08 | |
| Hailuo 2.3 (1080P) | MiniMax | $0.0817 |
| Gemini Omni Flash (720p, token-billed estimate) | $0.10 | |
| Kling 3.0 Pro (audio off) | fal.ai | $0.112 |
| Veo 3.1 Fast (audio) | $0.12 | |
| Wan 3.0 (1080P) | Alibaba Cloud | $0.20 |
| Ray3.2 (1080p) | Luma | $0.24 |
| FLUX 3 Video (1080p) | fal.ai | $0.29 |
| Veo 3.1 (audio) | $0.40 | |
| Seedance 2.0 (1080p) | Runway | $0.40 |
| Seedance 2.0 (1080p) | fal.ai | $0.682 |
| Seedance 2.5 (1080p) | fal.ai | $1.164 |
These rows are not all the same kind of output. Resolution, audio and token-based billing all change the comparison, so read the next sections before ranking models.
What audio adds
Some listings include audio in the price and some do not.
- Kling 3.0 Pro on fal.ai is $0.168 per second with audio and $0.112 without.
- Kling 3.0 Pro on Replicate is $0.336 per second with audio and $0.224 without.
- Veo 3.1 on Google Vertex AI is $0.2 per second, video only. Google's own listing is $0.4 per second with audio included. These are different listings, so the gap is not a clean audio premium.
If you do not need sound, look for a video-only SKU. If you do, confirm the listing says audio is included.
Same model, different providers
- Veo 3.1 at full quality is $0.4 per second on Google, fal.ai, Replicate, WaveSpeed and Runway, all with audio.
- Veo 3.1 Fast is $0.12 per second on Google and $0.15 on fal.ai, Replicate and WaveSpeed.
- Wan 3.0 at 1080p is $0.2 per second on Alibaba Cloud, fal.ai, WaveSpeed and Runway.
- Seedance 2.0 at 1080p runs from $0.4 per second on Runway to $0.682 on fal.ai. Replicate lists $0.45 and WaveSpeed lists $0.6.
- MiniMax H3 at 768p is $0.06 per second on fal.ai and $0.08 on Replicate and MiniMax's own API.
The maker's own API is not always the cheapest route. For MiniMax H3, fal.ai is cheaper than MiniMax itself. For Seedance 2.0, the spread between the cheapest and most expensive 1080p listing is large.
Seedance: the widest spread
Seedance 2.5 on fal.ai at 1080p is $1.164 per second. WaveSpeed lists $0.9 per second at 1080p, and Runway lists $0.68.
Replicate lists Seedance 2.5 at $0.2312 per second, but that is a 720p price with no video input, so it is not comparable to the 1080p rows. Compare resolution and input settings before you compare totals.
Retries multiply the bill
A list price is the cost of one render. In practice you pay for each attempt, and I have no data on whether providers bill failed jobs, so check that before you commit.
The formula is simple: clip length in seconds, times the per-second price, times the number of attempts you need. For example, Hailuo 2.3 is listed at $0.49 for a 6-second 1080P clip. That is the price of one take, and the lowest per-second price only wins if the first take is good. For a production pipeline, budget for several attempts per usable shot.
To estimate your own bill, use the TaskCents calculator. For how I measure text task costs, see the methodology.
What I did not measure
I have no observed cost per minute, retry rates or quality scores for video. Real bills can be lower with discounts or promotions, and higher once you count retries or longer prompts. Some listings, such as Vidu Q4 and MiniMax H3 Max on fal.ai, are noted as list price after a promo, so check the current price before you plan a budget.
FAQ
What is the cheapest AI video model?
In my data, Gen-4 Turbo on Runway is lowest at $0.05 per second, followed by MiniMax H3 at 768p on fal.ai at $0.06 per second. Compare output quality and resolution before choosing on price alone.
Does audio change the cost?
For Kling 3.0 Pro, yes: audio on costs more than audio off on both fal.ai ($0.168 versus $0.112 per second) and Replicate ($0.336 versus $0.224 per second). For Veo 3.1, the Vertex AI video-only listing is cheaper than Google's with-audio listing, but they are different listings, so check the exact SKU.
Why do my video costs run higher than the list price?
You pay for every attempt, not just the clip you keep. Your effective cost is the list price multiplied by the number of renders you need, so plan for retries.