For quality per dollar, Veo 3.1 Fast won our side-by-side test and costs $0.10 to $0.12 a second with audio on Vertex AI. For the cheapest clips, Veo 3.1 Lite and Seedance's Fast and Mini models run a few cents a second at 720p or below. Runway and Kling sell credits and units that convert to roughly $0.05 to $0.17 a second, depending on model and audio. OpenAI's Sora API is gone: it was shut down on September 24, 2026. For video you build from code, like slides, captions and trailers, Remotion is free for teams of up to three people.
Most prices are per second of generated video, and audio usually costs extra. Seedance is priced in tokens, so we show BytePlus's own per-second examples beside the token rate.
On episode 1, Veo 3 was the model everyone wanted and the most painful to pay for:
[Veo 3] right now is, I would say, like, pretty expensive. It's about $6 per eight second generation, and the generations are not perfect.
— Jordan Metzner, Episode 1, at 06:42
In the case of something like [Veo 3], sometimes takes up to ten minutes to generate like an eight second video
— Jordan Metzner, Episode 1, at 11:30
So in the live demo he picked the older model: "I would do a [Veo 3] but it's gonna take like ten minutes right now" (at 12:39). For the compliance-training videos built later on Veo 3.1 Fast, Sam said "60 to 70% of the content was right right off the bat" (at 08:27). Veo still mangles on-screen text: Halloween clips came out with "not like real words" (at 10:31).
Jordan built a standalone tool on the Sora 2 API the week it launched:
When I add a reference image, it has to be this exact size. Otherwise, it doesn't work. And sometimes, like, if you make too long of a of a post of of of an input, then sometimes the the the API fails
— Jordan Metzner, Episode 17, at 03:34
The clips themselves had problems too: "sometimes the the physics have been a bit weird… I had like three fingers or something" (at 11:57), and Sam found it "pretty wonky compared to to [Veo 3]" (at 11:11). After running the same prompt through both, Jordan said he'd fold Sora into the main studio "and, like, kind of deprecate the other one" (at 06:24).
We did a quick side by side test. Veo 3.1 looks better
— Sam Nadler, Episode 17, at 08:34
Sam's candidate highlight-reel builder renders video from code with Remotion:
one of the most challenging parts was to get the video to render in a reasonable amount of time. It still takes probably five to ten minutes
— Sam Nadler, Episode 36, at 07:13
Our only Runway and Kling experience is a cost decision: giving everyone accounts at "RunwayML and [Kling]" was the per-seat bill Jordan built the studio to avoid (at 18:54). We have no build of our own on those APIs or on Seedance's API, so they appear in the price table only. Jordan did note that Seedance generates in "just a few moments" where Veo 3 took minutes (at 11:30).
No. OpenAI's Sora 2 model page says the Sora API was shut down on September 24, 2026, and that there is no one-to-one replacement API. It had been priced at $0.10 per second.
On Vertex AI on November 21, 2026, Veo 3.1 was $0.40 with audio or $0.20 video-only at 720p to 1080p, Veo 3.1 Fast $0.10 to $0.12 with audio, and Veo 3.1 Lite $0.05 to $0.08 with audio. Google lists these per '1 count'. The Veo 3.1 preview models in the Gemini API were shut down on October 22, 2026.
Runway sells API credits at $0.01 each. Gen-4 Turbo uses 5 credits per second ($0.05), Gen-4.5 uses 12 ($0.12), and Veo 3.1 through Runway uses 40 credits per second with audio or 20 without.
When we ran the same prompt through both on the show, both hosts preferred Veo 3.1 Fast. Jordan found Sora 2's API fussy about reference image sizes and long prompts, and Sora 2 clips had odd physics and extra fingers.
Of the prices we checked on November 21, 2026, the lowest per second were Veo 3.1 Lite (video-only from $0.03 at 720p) and Seedance 2.0 Mini on BytePlus (about $0.04 a second at 480p in BytePlus's own example). Cheaper tiers usually cap resolution at 720p.