RenderDrama

What a second of AI video costs, channel by channel

What the resolution step costs on Sora

Sora publishes ten cents a second at 720p on the base model and thirty on the pro model at the same resolution, rising to seventy at 1080p. A batch tier halves all of them, which makes the range wider than resolution alone would. As of 2026-09-22.

Two decisions multiplying to seven timesTen cents a second at 720p on the base model, thirty on the pro model at the same resolution, seventy at 1080p on the pro model, and a batch tier that halves all of them. Model choice and resolution choice stack, so the top of the grid is seven times the bottom.Base at 720pTen cents asecond, and thefigure acomparison willquote.Change the modelPro at 720pThirty cents asecond, with nochange inoutput size atall.Change theresolutionPro at 1080pSeventy cents,the dearestunconditionalsecond in thisregister.Apply the batchtierHalvedFive tothirty-fivecents, appliedon every modeland resolutionon the page.
Fig. 1 A shortlist built on the ten-cent figure and delivered at the top combination is wrong by a factor, and both numbers sit in the same table.
Sora's published figures by model and resolution, and the modifier that applies to all of them. Read from API price page on 2026-09-22.
Model and tierPer second, as published
sora-2 at 720pTen cents
sora-2-pro at 720pThirty cents
sora-2-pro at 1080pSeventy cents
The batch tierHalf the standard figure, on every model and resolution

Inclusion rule. Figures on the API price page, by model and resolution, plus any modifier the page applies across them. The dearest published second in this register is on this page. Order. Ascending by published figure, with the modifier last.

1Where Sora sits in this column

The size of the step between resolutionsWhat each channel charges for the step from its lower resolution to its higher one, Sora marked. The ratio is the channel's own two figures divided, nothing more. Recorded 2026-09-22.The size of the step between resolutionsStep up in resolutionStep up inresolutionKling AIKling AI — Step up in resolution: Double to 1080p, five times to 4KMiniMaxMiniMax — Step up in resolution: Half again from 768P to 2KReplicateReplicate — Step up in resolution: Nearly three times for one stepSoraSora — Step up in resolution: Seven times from the cheapest to the dearestVeoVeo — Step up in resolution: Free on one tier, twelvefold across the rangeViduVidu — Step up in resolution: Just over double across three tiersWanWan — Step up in resolution: Four times from 480P to 1080P
Fig. 2 What each channel charges for the step from its lower resolution to its higher one, Sora marked. The ratio is the channel's own two figures divided, nothing more. Recorded 2026-09-22.
Step up in resolution, as each channel answering it has it, with Sora marked. Recorded 2026-09-22.
ChannelStep up in resolution
Kling AIDouble to 1080p, five times to 4K
MiniMaxHalf again from 768P to 2K
ReplicateNearly three times for one step
SoraSeven times from the cheapest to the dearest
VeoFree on one tier, twelvefold across the range
ViduJust over double across three tiers
WanFour times from 480P to 1080P

Inclusion rule. Channels publishing a rate at more than one resolution on the same page. A channel with a single figure has no step to price. Order. Alphabetical by channel name.

2The pro model costs three times as much at the same resolution

Thirty cents against ten at 720p is the whole difference between two models, with no change in output size. Then 1080p on the pro model adds another one hundred and thirty per cent, which is the steepest single resolution step on the ledger.

Stacked, those two decisions multiply to seven times. A shortlist built on the ten-cent figure and delivered at the top combination is wrong by a factor, and both figures are on the same page in the same table.

3A batch tier at half price changes what the ceiling means

Because the halving applies to every model and resolution rather than to one tier, the top of this page is thirty-five cents a second for work that can wait, not seventy. That is the same discount structure Vidu applies to off-peak hours, expressed as a queue rather than a clock.

For a serial production most generation is not urgent. Character sheets, scene plates and first passes can all queue, which makes the batch figure the realistic planning rate and the standard figure the cost of impatience.

4The dearest second in the register is here, and so is a cheap one

Seventy cents a second is the highest unconditional published figure anywhere on this ledger, and ten cents on the same page is unremarkable. A channel cannot usefully be ranked as dear or cheap when its own range covers that much ground.

Which is the argument for ranking rows rather than channels throughout this register. The model and the tier decide the rate; the shop mostly decides which models are on offer and what conditions come attached.

  • sora-2 at 720p
    $0.10 per secondOpenAI, API prices / recorded 2026-09-12
  • sora-2-pro at 720p
    $0.30 per secondOpenAI, API prices / recorded 2026-09-12
  • sora-2-pro at 1080p
    $0.70 per secondthe dearest published second in this ledgerOpenAI, API prices / recorded 2026-09-12
  • The batch tier
    Half the standard figureon every Sora model and resolutionOpenAI, API prices / recorded 2026-09-12

5Inside a generated second

Price aside, what goes into the seconds behind “Sora resolution step” is described in SceneMixer's guides:

The shots add up to more than the selected video tier allows for one segment.

SceneMixer, Shot Seconds: Changing Timing, and the Limits — “What does 'over cap' mean on the segment card?” (read 2026-10-13)

Every segment it can revise starts ticked; untick any of them, or press Clear and pick.

SceneMixer, Revise the Whole Episode: One Request, Every Segment — “Can I leave some segments alone?” (read 2026-10-13)

6Sources

Read from the API price page published at developers.openai.com on 2026-09-22. The rate this column sits beside is on Sora; what the register accepts as published is on the method page. The same column on other channels: Wan, Kling AI. Every column: what each channel publishes.