The 8 cheapest AI video models per second, from $0.015

The cheapest generally available AI video is now $0.015 a second. Pruna's P-Video-2-Pro-Reference is the low mark on a market where the price of a second of generated footage fell again this year, and where Google's Veo 3.1 family left the Gemini API today.

Every price below is the vendor's own published list rate for the cheapest generally available tier of video output, read from the vendor's own page on 22 October 2026 and converted to US dollars per second of finished video. Limited-time launch promotions are excluded from the ranking and noted in the entry they belong to. Where a platform resells another company's model, the entry says whose rate it is and which page this writer read. No hands-on tests were run for this list: it ranks on documented list price per second and nothing else. Google's Veo 3.1 tiers, whose $0.05-a-second Lite rate still circulates, no longer carry a buyable price on Google's own page and are handled in the Gemini entry below. In all, eight models made the cut.

8. Seedance 2.0 and 2.5 (ByteDance, on Runway)

Runway sells ByteDance's Seedance family on both its creative app and its developer API, and its own developer price page is where these rates live. At Runway's stated $0.01 per credit, Seedance 2.0 Mini runs $0.16 a second at 480p or 720p, Seedance 2.5 runs $0.20 a second at 480p, and plain Seedance 2.0 runs $0.36 a second. Every one of those is Runway's resale rate for ByteDance's model, not a rate ByteDance publishes itself.

Seedance 2.5 also bills its input and reference video by the second on top of the output, which can push a real request above the headline number. It tops out this list because the same motion quality that makes it a favorite for long shots is charged at the top of the market here. If you want Seedance motion at the lowest price, the Mini tier is the one to price out.

7. Gemini Omni Flash (Google)

Google's current video model in the Gemini API is Gemini Omni Flash, and its rate is published in tokens rather than seconds. Google's pricing page lists the video output at $17.50 per million output tokens, billed at 5,792 tokens per second of 720p video, which the page states works out to an effective price of approximately $0.10 a second. Google's deprecation schedule names the generally available model as gemini-omni-1.1-flash.

The Veo 3.1 generation is gone from that page. Google's deprecation schedule lists veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview all shutting down on 22 October 2026, each replaced by gemini-omni-1.1-flash. So the $0.05-a-second Veo 3.1 Lite figure that comparison tables keep printing is no longer a price a developer can buy on Google's own page, and the same page now carries no Veo price at all. One reseller beat Google on its own model: Runway lists Gemini Omni Flash 1.1 at 3.4 credits a second, or $0.034, at 360p, which is Runway's rate and not Google's.

6. Kling 3.0 (Kuaishou, on Runway)

Kling 3.0 is Kuaishou's model, and Runway carries it on its developer API. Runway's developer price page lists the Standard tier at 9 credits a second without audio, which at $0.01 a credit is $0.09 a second. The Pro tier is 12 credits, or $0.12 a second, without audio, and the 4K tier is 43 credits, or $0.43 a second.

Audio is the swing factor on Kling 3.0. Standard with audio runs 13 credits a second, or $0.13, and Pro with audio runs 17 credits, or $0.17, and Runway's page notes that the no-audio rate applies only when audio is turned off. The 4K tier is billed the same either way. All of these are Runway's posted rates for Kuaishou's model, read from Runway's own page on 22 October 2026.

5. LTX-2.5 (Lightricks)

LTX-2.5 is Lightricks' own model, and Lightricks prices it on its own API documentation. The Fast variant runs $0.09 a second at 720p, $0.13 at 1080p, $0.19 at 1440p and $0.30 at 4K. The Pro variant runs $0.12 a second at 720p, rising to $0.39 at 4K. Billing is per second of output, with no request fee or per-asset charge, and the same rates cover text-to-video, image-to-video and audio-to-video.

The $0.09 a second Fast rate at 720p is the cheapest way to get LTX-2.5 output, and it is a real step down from the $0.12 figure that gets quoted for the Pro tier. LTX-2.5 is also available as open weights, so the per-second price only applies if you would rather pay an API than run the model yourself. Runway carries LTX as well, but Lightricks' own page is the source for the numbers here.

4. MiniMax H3 (MiniMax)

MiniMax sells its H3 video model on its own platform, and its pricing overview lists the rates directly. MiniMax-H3 runs $0.08 a second at 768P and $0.13 a second at 2K. MiniMax-H3-Max runs $0.05 a second at 480P and $0.08 a second at 768P. That makes $0.05 the cheapest generally available tier in the H3 family, and it is the number the family is ranked on here.

The rates are per second of output, and input material is billed separately: the first five images are free on MiniMax-H3 and each extra image is $0.04, while reference video is billed at the output resolution's per-second rate. Audio is free on both models. MiniMax's page also notes that its H3 video packages do not yet support the H3 model, so H3 runs on the pay-as-you-go API. Runway resells H3 Max at 5 credits a second at 480p and 8 credits at 768p, its own rate that matches MiniMax's.

3. Vidu Q4 Preview (Shengshu)

Shengshu's Vidu Q4 Preview is priced in credits on the Vidu API platform, at a standard rate of $0.005 per credit stated on the platform's own recharge page. Q4 Preview burns 9 credits a second at 540p, 19 at 720p, 24 at 1080p, 38 at 2K and 78 at 4K, which converts to $0.045 a second at 540p and $0.39 a second at 4K. Those two figures match the ones in Shengshu's GlobeNewswire release, which states pricing starting at $0.045 a second at 540P and scaling to $0.39 at 4K.

The lower number that circulated at launch, $0.014 a second, is not this list price. It matches the consumer Vidu app's promotional 540p rate of 4 credits a second on the free tier, not the API's own $0.005-per-credit list rate, so it belongs to a subscription promotion rather than the rate a developer buys. Shengshu's release also states that image-to-video and reference-to-video generation carry a 30% discount through November 30, 2026, which is a promotion and stays out of the ranking. Q4 Preview outputs up to 4K with 10-bit color, which is why the top of its price range runs so high.

2. HeyGen Video (HeyGen)

HeyGen Video is HeyGen's own general-purpose video model, and its list price is stated on HeyGen's own pages. HeyGen's developer catalog lists the model at $0.03 a second at 768p with audio as its list price before launch promotions, and HeyGen's own price guide states $0.02 a second at 480p and $0.03 at 768p with audio, with 1080p and 2K at three times the 768p rate. The $0.02 at 480p is the cheapest generally available tier, which is where the model sits on this list.

HeyGen Video launched on September 30, 2026 and is built on MiniMax's H3 model, post-trained by HeyGen, so the base weights come from the same family as the entry above. Every clip comes back with native audio generated in the same call. The one caveat is timing: HeyGen's own page states launch pricing is half off through October 31, 2026, so the discounted rate a buyer sees today is a promotion, and the list rate above is what applies from November.

1. Pruna P-Video-2-Pro-Reference (Pruna AI)

Pruna's P-Video-2-Pro-Reference holds the lowest list rate on the market, and it is published in Pruna's own documentation. In cost mode the model runs $0.015 a second of 480p output and $0.025 at 768p, against $0.02 and $0.035 a second in the default speed mode for the same quality. That $0.015 a second is the figure this list is ranked on, and it is the only rate on the list below two cents a second.

The catch is what the model is. It is reference-conditioned and based on MiniMax's H3 model, so a prompt alone is rejected and every request needs at least one image or video reference. References are billed on top: $0.005 per image, $0.03 a second of reference video at 480p and $0.0015 a second of reference audio in cost mode, which brings the cheapest possible request to $0.075. For anyone generating fresh footage from a reference, that is still the lowest per-second rate on the board, and Pruna's own page states the cost mode matches the speed mode's output quality.