Original research

AI Video Model Price Index (October 2026)

Updated Oct 1, 2026 · By Zachary Warren, Founder & CEO

Key findings

1. Cheapest per second

   The cheapest 720p video with native audio comes from Veo 3.1 Lite on the Gemini API at $0.05 per second, or $0.40 for an 8-second clip. The lowest per-second price overall is $0.03, for Veo 3.1 Lite on Vertex AI (720p, no audio), and Luma Ray3.2 sells 5-second 540p clips for $0.15 each.

2. What a typical 8-second clip costs

   Across 48 live price points sold directly by model makers, the median 8-second clip costs $0.93, and 27 of the 48 cost less than $1.00. The range runs from $0.24 (Veo 3.1 Lite on Vertex AI, 720p, no audio) to $6.22 (Seedance 2.0 on BytePlus ModelArk, 4K, with audio).

3. Marketplace vs official price

   Seedance 2.0, Seedance 2.0 Fast and Seedance 2.5 are priced 82% to 114% higher per second on fal.ai than on BytePlus ModelArk, across 5 matching tiers. Kling 3.0 is priced identically on fal.ai and on Kling's official API: the same per-second rate at all 5 matching tiers, from $0.084 to $0.42.

4. The 4K premium

   For the same model on the same API, 4K costs 50% to 150% more per second than 1080p across the 7 official price pairs that list both. The smallest jump is Veo 3.1 on the Gemini API, from $0.40 to $0.60 per second; the largest, +150%, is for Veo 3.1 Fast and Kling 3.0.

5. Sora 2 is retired

   Sora 2 is no longer available: OpenAI's Sora API was retired on September 24, 2026. Its last list price was $0.10 per second at 720p with audio, and Sora 2 Pro listed $0.30 to $0.70. The nearest-priced live 720p models with native audio are Veo 3.1 Fast ($0.10/s) and Gemini Omni Flash ($0.10136/s).

Every price in the index

Prices are list prices in USD before tax, volume discounts, promotions or subscriptions. "—" = not published by the provider. 8-second and 30-second costs are output seconds × the per-second price; 30 seconds may need several clips. Row numbers link to each price's source note below.

Same model, different channel

Some models are sold both by their maker and by fal.ai, a third-party inference marketplace. Matching the same model, resolution and audio setting:

* fal.ai does not label the resolution of these Kling tiers; they are matched to Kling's official 720p and 1080p rates because the per-second prices are identical.

What these models cost through UGC Copilot

Credits for one 8-second clip on each engine UGC Copilot renders with, priced at the Pro plan's included-credit rate ($79 for 1,800 credits a month, $0.0439 per credit) and at the 200-credit pack rate ($25, $0.125 per credit).

These are not apples-to-apples with the raw API prices above. UGC Copilot credits also pay for the workflow around a render: the script (1 credit), the AI persona and scene images (from 1 credit each), one free automatic re-render when a clip fails the post-render quality check, and an automatic refund when a render fails.

Seedance 2.5 (Ultra) is at launch pricing through Oct 31.

See UGC Copilot plans and credit packs · Compare the engines side by side

Methodology and caveats

Every price was read on 2026-10-01 from the provider's own pricing/docs page or, where the public API is sold through an inference marketplace, from that marketplace's official model page (channel='marketplace: fal.ai'). Static pages were fetched as raw HTML; JS-rendered pages (kling.ai/dev/pricing, BytePlus ModelArk pricing, Google Cloud Vertex pricing) were rendered in a browser and their visible text read. Nothing was filled from memory; numbers absent from an official page are null with the reason in notes. Credit/unit/token prices are converted to USD per second using the conversion stated on the same page (shown in notes). costFor8sUsd / costFor30sUsd = pricePerSecondUsd x 8 / x 30, rounded half-up to cents, computed only when pricePerSecondUsd is known; they are cost of that many seconds of generated output and ignore single-clip maximums (30 s may require several clips), per-request minimums, input-media fees and prompt-token charges. List prices in USD before tax, volume discounts, promos or subscriptions unless noted.

• Channel: 'official' = model maker's own API; 'marketplace: fal.ai' = third-party inference marketplace's official model page.
• Status: 'live' unless stated.

Caveats

• Vertex AI lists Veo 3.1 prices per '1 count' without defining the unit. We treat it as per second because the same page's with-audio rows exactly match the Gemini API's explicit per-second prices.
• fal.ai's Seedance 2.0 page shows two slightly different 720p per-second prices (its docs table and its playground blurb); we use the docs-table figure. fal.ai's Seedance 2.5 per-second prices are its own approximations; its token formula is authoritative.
• BytePlus ModelArk bills Seedance in tokens. The per-second figures here are the page's own price example for a 5-second 16:9 video divided by its length, so other aspect ratios, and any input video, change the cost. A limited-time enterprise discount on Seedance 2.0 Fast is not applied.
• fal.ai does not label the resolution of its Kling 3.0 Standard and Pro tiers. We list them as 720p and 1080p because their per-second prices equal Kling's official 720p and 1080p rates, so the channel match for those tiers rests on price.
• Kling's official API is sold as prepaid unit packages; the per-second figures use the list unit price, and larger packages are discounted (see the source notes).
• Luma prices Ray3.2 per clip, not per second, so the per-second, 8-second and 30-second columns are blank for it. Its FAQ says billing is per 5-second block, which contradicts its own table: a 10-second clip costs 3× the 5-second price at every resolution. We report the table.
• Pika's API requires a paid monthly membership on top of pay-as-you-go usage, and its listed prices already include Pika's platform fee.
• Where a provider's pricing and API docs do not say whether audio is generated (Runway, Luma AI, MiniMax and Pika), native audio is shown as "Not stated" rather than assumed.
• Gemini Omni Flash is billed in output tokens; its per-second figures use the tokens-per-second rate stated on the pricing page and exclude input tokens.
• Sora 2 and Sora 2 Pro are shown at the last price OpenAI published, which its model pages still display, in a separate Retired group.

How to cite this

UGC Copilot (2026). AI Video Model Price Index (October 2026). Prices verified October 1, 2026. https://ugccopilot.ai/research/ai-video-model-price-index/ (data: https://ugccopilot.ai/research/ai-video-model-price-index-2026-10.csv, CC BY 4.0).

The data is free to reuse under CC BY 4.0 with attribution and a link to this page. Download ai-video-model-price-index-2026-10.csv.