Wednesday, 18 November 2026 | Updating Daily AI insight, written for builders

AI Video Generation Pricing 2026: Gemini Omni, Veo 3.1, Kling and More per Second

Updated · Originally published
  • Google shut down the Veo 3.1 preview models in the Gemini API on 22 October 2026 and now points developers to Gemini Omni Flash. Gemini Omni 1.2 Flash, released on 5 November 2026, costs about $0.07 per second of 720p video and $0.023 at 360p.
  • Veo 3.1 itself is still on sale elsewhere: Google Cloud’s Gemini Enterprise Agent Platform lists the same figures the retired previews charged per second ($0.40 with audio at 720p/1080p, $0.60 at 4K). fal, which charged the same, deprecated its Veo 3.1 endpoint on 17 November 2026.
  • Through fal, MiniMax H3 and Wan 2.5 cost $0.05 per second at 480p, Kling 2.5 Turbo Pro $0.07, Kling 4.0 Flash $0.094 and Kling 4.0 $0.157 at 720p.
  • Sora 2 is no longer available: OpenAI shut down the Sora 2 API and the sora-2 and sora-2-pro models on 24 September 2026, with no replacement. Its old rates are kept below for reference only.
  • A 30-second clip costs about $2.09 on Gemini Omni 1.2 Flash (720p), $4.71 on Kling 4.0 (720p) and $12.00 on Veo 3.1 with audio (720p/1080p).

AI video generation cost is quoted per second of output rather than per request, which makes it unusually easy to compare across providers and unusually easy to underestimate. A rate that looks trivial at $0.40 becomes $24 for a single one-minute clip, and a workflow that regenerates a shot ten times before it is usable multiplies that again. Every figure below comes from the provider’s own pricing page or, for models sold through an aggregator, from fal’s model page, checked on 6 November 2026. Sora 2 rates and the Veo 3.1 Gemini API preview rates are historical.

Quick answer: what is the cheapest AI video generation API per second?

The cheapest published AI video generation rate in November 2026 is Google’s Gemini Omni 1.2 Flash, at $0.023 per second of 360p video (about $0.07 at 720p, with native audio). At 720p, Google Cloud also lists Veo 3.1 Lite, a preview model there, at $0.03 per second without audio and $0.05 with it. Through fal, MiniMax H3 and Wan 2.5 cost $0.05 per second at 480p, and Kling 2.5 Turbo Pro $0.07. At those rates a 30-second clip costs $1.50 to $2.10 per successful generation. At the top end, Seedance 2.5 reaches about $1.16 per second at 1080p on fal. One caveat before comparing: a usable shot can take several attempts, so multiply any rate by your real retry count.

AI video pricing: cost per second, compared

For subscription plans as well as APIs, our AI Video Generator Cost Index converts each price into the cost of a 5-second 1080p clip; the data is free to download.

Model Resolution Price / second Source
Gemini Omni 1.2 Flash 360p $0.023 Google
Veo 3.1 Lite (video only, preview) 720p $0.03 Google Cloud
MiniMax H3 480p $0.05 fal
Wan 2.5 480p $0.05 fal
Veo 3.1 Lite (with audio, preview) 720p $0.05 Google Cloud
MiniMax H3 768p $0.06 fal
Gemini Omni 1.2 Flash 720p ~$0.07 Google
Kling 2.5 Turbo Pro — $0.07 fal
Kling 3.0 Standard (audio off / on) — $0.084 / $0.126 fal
Kling 4.0 Flash 720p $0.094 fal
Wan 2.5 720p / 1080p $0.10 / $0.15 fal
Veo 3.1 Fast (with audio) 720p / 1080p $0.10 / $0.12 Google Cloud
Luma Ray3.2 540p / 720p / 1080p $0.10 / $0.20 / $0.40 fal
Gemini Omni 1.2 Flash 1080p (upsampled) $0.104 Google
MiniMax H3 2K / 4K $0.13 / $0.16 fal
Kling 4.0 720p / 1080p $0.157 / $0.235 fal
Gemini Omni 1.2 Flash 4K (upsampled) $0.209 Google
Seedance 2.5 (US-hosted) 480p / 720p / 1080p ~$0.22 / ~$0.47 / ~$1.16 fal
Veo 3.1 (with audio) 720p / 1080p / 4K $0.40 / $0.40 / $0.60 Google Cloud

At 720p with sound, published rates run from $0.05 (Veo 3.1 Lite, a preview on Google Cloud) and about $0.07 (Gemini Omni 1.2 Flash) to $0.40 for Veo 3.1 and about $0.47 for Seedance 2.5 on fal: close to a tenfold spread at the same resolution. That gap is far wider than the quality gap most projects would notice, which is the single most useful thing to know before picking a default.

Shut down: rates kept for reference

Model Resolution Last price / second Status
Veo 3.1 Lite (Gemini API preview) 720p / 1080p $0.05 / $0.08 Shut down 22 Oct 2026
Veo 3.1 Fast (Gemini API preview) 720p / 1080p / 4K $0.10 / $0.12 / $0.30 Shut down 22 Oct 2026
Veo 3.1 Standard (Gemini API preview) 720p, 1080p / 4K $0.40 / $0.60 Shut down 22 Oct 2026
Sora 2 720p $0.10 OpenAI (shut down 24 Sep 2026)
Sora 2 Pro 720p / 1024p / 1080p $0.30 / $0.50 / $0.70 OpenAI (shut down 24 Sep 2026)

What a real clip actually costs

Model 10 seconds 30 seconds 60 seconds
MiniMax H3 (768p) $0.60 $1.80 $3.60
Gemini Omni 1.2 Flash (720p) ~$0.70 ~$2.09 ~$4.17
Kling 2.5 Turbo Pro $0.70 $2.10 $4.20
Kling 4.0 Flash (720p) $0.94 $2.82 $5.64
Kling 4.0 (720p) $1.57 $4.71 $9.42
Luma Ray3.2 (720p) $2.00 $6.00 $12.00
Veo 3.1 with audio (720p/1080p) $4.00 $12.00 $24.00
Seedance 2.5 (720p) ~$4.73 ~$14.19 ~$28.38

These are the cost of one successful generation. Video models are not one-shot tools: a usable shot can take several attempts as you adjust the prompt, seed or reference image. Most models also cap a single generation at 8 to 15 seconds (Kling 4.0 and Seedance 2.5 reach 30), so a minute of video always means several generations or extensions. Budget the table above multiplied by your real retry count, and the difference between a $0.05 and a $0.40 model stops being academic.

Gemini Omni Flash and Veo 3.1 pricing, and the Veo 3 trap

Google’s Gemini API now sells video generation through Gemini Omni Flash. Omni 1.2 Flash (gemini-omni-1.2-flash) is billed by output tokens: $12 per million video tokens, with 720p video counted at 5,792 tokens per second, which Google works out at about $0.07 per second (about $0.70 for a 10-second clip). The same footnote gives $0.023 per second at 360p, $0.104 at 1080p and $0.209 at 4K, the last two upsampled. Clips run 3 to 15 seconds with native audio. The previous version, Omni 1.1 Flash, bills video at $17.50 per million tokens, about $0.10 per second, and Google will route its requests to 1.2 Flash, at 1.2 rates, from 3 December 2026. Neither has a free tier or a batch rate.

The Veo 3.1 Preview models in the Gemini API were shut down on 22 October 2026. Veo 3.1 continues on Google Cloud’s Gemini Enterprise Agent Platform, where veo-3.1-generate-001 and veo-3.1-fast-generate-001 are generally available (4-, 6- or 8-second clips, up to 4K on the standard model, with retirement no earlier than 4 May 2027). Its pricing page lists the same figures the previews had: Standard $0.40 with audio at 720p/1080p and $0.60 at 4K ($0.20 and $0.40 without audio), Fast from $0.10 with audio ($0.08 without), and Lite, still a preview there, from $0.05 ($0.03 without audio). fal, which sold Veo 3.1 at $0.40 per second with audio, deprecated its endpoint on 17 November 2026.

Google shut Veo 3 down in the Gemini API on 30 June 2026. This matters because Veo 3 pricing is still one of the most-searched terms in this category, and much of the content answering it describes a model the Gemini API no longer runs. If you are choosing now, price Gemini Omni 1.2 Flash or Veo 3.1 on Google Cloud.

The tier structure is the useful part. Inside Veo 3.1, Lite costs an eighth of Standard at the same 720p resolution, and inside Omni 1.2 Flash, 4K costs about nine times as much as 360p. The tier choice, not the vendor choice, is usually where the money goes.

Kling pricing: 4.0, 4.0 Flash, 3.0 and 2.5 Turbo

Through fal, Kling 4.0 is $0.157 per second at 720p and $0.235 at 1080p, for clips of 3 to 30 seconds, and Kling 4.0 Flash is $0.094 per second at 720p, for clips of up to 20 seconds. Kling says Kling 4.0 rolled out to all users on 15 October 2026. The older generations are cheaper: Kling 3.0 Standard is $0.084 per second with audio off and $0.126 with audio on (Pro: $0.112 and $0.168), and Kling 2.5 Turbo Pro is $0.07 per second ($0.35 for the first five seconds, then $0.07 for each extra second). Rates for models routed through an aggregator can differ from the vendor’s own console, so treat these as that aggregator’s rates rather than universal ones.

The cheaper alternatives: MiniMax H3 and Wan

Through fal, MiniMax H3 is $0.05 per second at 480p, $0.06 at 768p, $0.13 at 2K and $0.16 at 4K, and Wan 2.5 is $0.05 per second at 480p ($0.10 at 720p, $0.15 at 1080p). Alibaba’s newer Wan 3.0 is listed at $0.05, $0.10 and $0.20 for the same three resolutions. MiniMax has also published H3’s weights on Hugging Face under a community license, so it can run on your own hardware as well.

For high-volume, lower-stakes work such as social clips, background footage and concept passes, the $0.05 to $0.07 band does the job at an eighth to a sixth of Veo 3.1’s $0.40 per second with audio.

Luma Ray3.2 and Seedance 2.5

Luma Ray3.2 is billed by duration and resolution on fal: a 5-second clip costs $0.50 at 540p, $1.00 at 720p and $2.00 at 1080p, which is $0.10, $0.20 and $0.40 per second. Seedance 2.5, ByteDance’s 30-second model, is token-billed on fal’s US-hosted endpoint; fal puts it at roughly $0.22 per second at 480p, $0.47 at 720p and $1.16 at 1080p, which makes it the most expensive per-second option here at 1080p.

Sora 2 pricing

OpenAI shut down the Sora 2 API, along with the sora-2 and sora-2-pro models, on 24 September 2026, and named no replacement, so the rates in this section are a historical record. OpenAI charged $0.10 per second for sora-2 at 720p, in either portrait (720×1280) or landscape (1280×720). The Pro tier was priced by resolution: $0.30 at 720p, $0.50 at 1024p and $0.70 at 1080p.

The detail most comparisons missed was the batch tier, which halved everything. sora-2 fell to $0.05 per second and sora-2-pro to $0.15 / $0.25 / $0.35. For non-interactive work, such as overnight renders, bulk asset production or anything queued, batch made Sora 2 the cheapest frontier video model available until the shutdown.

AI image generation pricing

Image pricing does not use the same unit, which makes direct comparison harder.

Model Price Unit
Nano Banana 2.1 $0.0336 (1K) / $0.0504 (2K) / $0.113 (4K) per image
gpt-image-2.6 (sunburst / flare) $8 in / $30 out per 1M image tokens
gpt-image-2.5 (sunburst / flare) $8 in / $30 out per 1M image tokens
Imagen 4 Fast / Standard / Ultra (Google Cloud; Gemini API shut down 17 Aug 2026) $0.02 / $0.04 / $0.06 per image
gpt-image-1 Shut down 23 Oct 2026 —
gpt-image-1.5 (shuts down 1 Dec 2026) $8 in / $32 out per 1M image tokens
gpt-image-1-mini (shuts down 1 Dec 2026) $2.50 in / $8 out per 1M image tokens

OpenAI bills images as tokens rather than per picture, so the cost of a single image depends on its size and detail instead of being a flat rate. That is harder to forecast, but cheaper for small outputs. Batch halves those token rates where it is offered ($4 in / $15 out for gpt-image-2.5; OpenAI’s pricing page lists no batch rate for gpt-image-2.6), and Google’s batch tier halves Nano Banana 2.1 to about $0.017 per 1K image.

Google shut Imagen 4 down in the Gemini API on 17 August 2026, and its deprecations page names Nano Banana 2.1 (gemini-nano-banana-2.1) as the replacement, at about $0.034 per 1K image; Google Cloud’s pricing page still lists Imagen 4. As with Veo 3, a lot of published Imagen 4 pricing now describes a model the Gemini API no longer runs. OpenAI is retiring its first GPT Image models the same way: gpt-image-1 shut down on 23 October 2026, with gpt-image-2.5-sunburst or gpt-image-2.5-flare as its recommended replacement, and gpt-image-1-mini and gpt-image-1.5 follow on 1 December 2026, with gpt-image-2.6-sunburst or gpt-image-2.6-flare as theirs.

How to choose without overpaying

Start at the bottom of the range, not the top. Most projects default to the flagship tier and never test whether a Lite, Fast or lower-resolution option is good enough for the shot they actually need. At an eight-fold price difference inside Veo 3.1 alone, that test is worth running before you scale.

Use batch wherever the work is not interactive. Halving the rate is a larger saving than most vendor switches, and it costs nothing but latency. For images, OpenAI and Google batch tiers still halve the rate; for video, the only batch tier priced here went with Sora 2, Google lists no batch rate for Gemini Omni Flash, and Veo 3.1 on Google Cloud does not support batch inference.

Count retries, not generations. The honest unit of cost is seconds of finished video: if a usable shot takes five attempts, you pay five times the sticker price.

Check the deprecation notice before you build. Several of the most-quoted models in this category have already gone from their original APIs: Veo 3 on 30 June 2026, Imagen 4 on 17 August 2026 and the Veo 3.1 previews on 22 October 2026 in the Gemini API, Sora 2 on 24 September 2026 and gpt-image-1 on 23 October 2026. Since then, fal deprecated its Veo 3.1 endpoint on 17 November 2026. OpenAI retires gpt-image-1-mini and gpt-image-1.5 on 1 December 2026, and Gemini Omni 1.1 Flash routes to 1.2 Flash from 3 December 2026.

If you are still choosing a model rather than a price, our best AI video generators guide ranks them by use case. If your workload also involves text models, the same per-unit discipline applies there. The AI API cost calculator works out monthly spend per model, and the AI models database carries specifications and pricing for every major model we track. For running generation locally instead, the VRAM calculator sizes the GPU you would need.

Related: Sora 2’s API was shut down on 24 September 2026, and image-to-video costs the same per second as text-to-video on every model priced here.

Prefer a subscription to an API? Ten AI video platforms with real prices, and which three have a genuinely usable free tier.

Frequently asked questions

What is the cheapest AI video generator API?

Among models with published rates in November 2026, Google’s Gemini Omni 1.2 Flash is the cheapest at $0.023 per second of 360p video, or about $0.07 at 720p with native audio. Google Cloud lists Veo 3.1 Lite, a preview there, at $0.03 per second at 720p without audio and $0.05 with it, and through fal MiniMax H3 and Wan 2.5 cost $0.05 per second at 480p, with Kling 2.5 Turbo Pro at $0.07. The Veo 3.1 Lite preview in the Gemini API, which also cost $0.05, was shut down on 22 October 2026.

Is Veo 3.1 still available in the Gemini API?

No. Google shut down veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview on 22 October 2026 and recommends Gemini Omni Flash (gemini-omni-1.2-flash) instead. Veo 3.1 and Veo 3.1 Fast remain generally available on Google Cloud’s Gemini Enterprise Agent Platform, whose pricing page lists the same figures the previews charged per second ($0.40 with audio at 720p/1080p, $0.60 at 4K); fal, which sold Veo 3.1 at the same rate, deprecated its endpoint on 17 November 2026.

How much does Kling 4.0 cost per second?

Through fal, Kling 4.0 costs $0.157 per second at 720p and $0.235 at 1080p, for clips of 3 to 30 seconds, so a 30-second 720p clip is $4.71 per successful generation. Kling 4.0 Flash is $0.094 per second at 720p. Kling says Kling 4.0 rolled out to all users on 15 October 2026.

What does a one-minute AI video cost?

For one successful generation: $3.60 on MiniMax H3 at 768p, about $4.17 on Gemini Omni 1.2 Flash at 720p, $4.20 on Kling 2.5 Turbo Pro, $9.42 on Kling 4.0 at 720p and $24.00 on Veo 3.1 with audio. Most models cap one generation at 8 to 15 seconds, so a minute means several generations or extensions, and you should multiply by your retry rate, because a usable shot rarely lands on the first attempt.

Is Veo 3 still available?

Not in the Gemini API. Google shut Veo 3 down there on 30 June 2026 and directs developers to Gemini Omni Flash (gemini-omni-1.2-flash) or the GA models on its Gemini Enterprise Agent Platform; the Veo 3.1 Preview models in the Gemini API also shut down on 22 October 2026. Much of the Veo 3 pricing content published online still describes the retired model, so check the tier and the version before budgeting against it.

How much does Sora 2 cost per second?

Sora 2 is no longer on sale: OpenAI shut down the Sora 2 API and the sora-2 and sora-2-pro models on 24 September 2026, with no replacement. Before the shutdown, Sora 2 cost $0.10 per second of 720p video on the standard tier and $0.05 per second on the batch tier, and Sora 2 Pro was $0.30 per second at 720p, $0.50 at 1024p and $0.70 at 1080p, halving to $0.15 / $0.25 / $0.35 on batch.

Is Veo cheaper than Sora?

Sora is no longer an option: OpenAI shut the Sora 2 API down on 24 September 2026. While both were on sale, it depended on the tier. Veo 3.1 Lite at $0.05 per second was cheaper than Sora 2 at $0.10, and Veo 3.1 Fast matched it at $0.10, but Veo 3.1 Standard at $0.40 per second was four times the price of Sora 2. Comparing vendors without naming the tier is meaningless in this category.

Why is AI video priced per second instead of per video?

Generation cost scales almost linearly with output length, because the model produces frames sequentially. Per-second billing therefore reflects the provider’s actual compute more honestly than a flat per-video fee would, and it is why a short clip is cheap while a one-minute render is not.

Prices are the published list rates from each provider, checked on 6 November 2026 against the Google Gemini API pricing page, the Google Cloud Agent Platform pricing page, the OpenAI API pricing page and fal’s model pages. Sora 2 rates and the Veo 3.1 Gemini API preview rates are historical. Aggregator rates can differ from a vendor’s own console, and all providers change pricing without notice.

Written by Mustafa Ihsan

Mustafa Ihsan is the founder and editor of Convly.ai. He built and maintains the site's live AI models database, its price-performance index, and its free calculators for VRAM requirements, API costs and self-hosting economics. He writes about model pricing, benchmark results and the hardware needed to run AI models locally, and consistently prefers measured numbers to vendor claims.

Scroll to Top