Sora Pricing (2026): API Costs Before Shutdown

Sora's app is gone and its API ends September 24. See current per-second prices, cost per usable clip, limits, refunds, and migration options.

Thursday, August 13, 2026Omid Saffari
Tools
Sora Pricing (2026): API Costs Before Shutdown

Sora no longer has a consumer subscription: its web and app experiences closed on April 26, 2026, and the API ends on September 24. Until then, API rates run from $0.10 to $0.70 per generated second, but the sensible budget for any continuing workflow is migration, not another Sora integration.

Sora pricing at a glance

Sora has no live monthly or annual consumer plan. New generation access is API-only until September 24, when OpenAI says the Sora API will close.

OpenAI Sora 2 model page showing current per-second video pricing
OpenAI Sora 2 Model Pricing

Every number below was verified against OpenAI's live Sora model and API pages on August 13, 2026. The eight-second column uses a duration the current Videos API supports, so it is a purchasable job rather than a rate-card fiction.

Sora accessAvailabilityCurrent price8-second output
Web and appDiscontinued April 26UnavailableUnavailable
Sora 2, 720pAPI until September 24$0.10/second$0.80
Sora 2 Pro, 720pAPI until September 24$0.30/second$2.40
Sora 2 Pro, 1024pAPI until September 24$0.50/second$4.00
Sora 2 Pro, 1080pAPI until September 24$0.70/second$5.60*

*OpenAI publishes the 1080p rate, but its public Videos API size schema does not currently list a 1080p value. Verify that option in your account before treating $5.60 as an executable job.

The billing unit is output time, not a seat or a monthly allowance. There is no annual discount, included pool, or subscription overage calculation because there is no current Sora subscription. API usage accrues at the published per-second rate.

The current API offers 4, 8, or 12-second durations. That makes native Sora 2 clips cost $0.40, $0.80, or $1.20. Sora 2 Pro costs $1.20, $2.40, or $3.60 at 720p; $2.00, $4.00, or $6.00 at 1024p; and a rate-card $2.80, $5.60, or $8.40 at 1080p.

A ten-second equivalent is easy to quote: $1.00 for Sora 2 and $3.00, $5.00, or $7.00 for Sora 2 Pro. It is not a selectable ten-second job. A ten-second deliverable needs editing around the 4, 8, and 12-second generation choices, which can make the invoice higher than a simple ten-times-rate calculation.

What Sora costs per usable clip

Your useful cost is the generation price multiplied by the number of attempts required to get a keeper. A keeper is the clip that survives review and reaches the edit, not every file the model returns.

Call that the retry multiplier. If one of every four generations is usable, the multiplier is four. This is a planning assumption, not a claimed Sora benchmark. Replace it with your own acceptance rate as soon as you have production data.

For one eight-second Sora 2 keeper, four attempts cost 4 × $0.80, or $3.20. At Sora 2 Pro's published 1080p rate, four attempts cost 4 × $5.60, or $22.40. The resolution choice changes the keeper budget by $19.20 before editing, upscaling, captions, or human review enters the workflow.

The same point is clearer per generated minute:

  • Sora 2 at 720p costs $6.00 per generated minute.
  • Sora 2 Pro at 720p costs $18.00 per generated minute.
  • Sora 2 Pro at 1024p costs $30.00 per generated minute.
  • Sora 2 Pro at its published 1080p rate costs $42.00 per generated minute.

At a four-attempt retry multiplier, one kept minute costs $24, $72, $120, or $168 before post-production. The cheap-looking $0.10 rate stays cheap only when 720p is acceptable and the keeper rate is healthy.

Take a DTC brand planning 30 final eight-second social cuts. Thirty clips equal 240 kept seconds. With four attempts per keeper, the campaign creates 960 generated seconds. Sora 2 costs $96. Sora 2 Pro at the published 1080p rate costs $672, subject to the public API schema mismatch. That $576 gap is a budget line, not a rounding error.

The cost model also explains why a lower-quality draft tier can be the smarter first pass. Generate concepts at $0.10 per second, reject weak motion and composition cheaply, and reserve Pro only for the shot where its model or resolution earns the multiplier. Paying Pro rates for all exploratory takes puts the most expensive compute at the least certain stage of the job.

Sora 2 versus Sora 2 Pro has no break-even point

Sora 2 Pro never becomes cheaper through volume. Every tier is linear per second, so Pro remains 3 times the Sora 2 rate at 720p, 5 times at 1024p, and 7 times at the published 1080p rate.

That means the tier decision does not turn on monthly usage. It turns on whether the higher model or resolution creates enough additional keepers, avoids enough post-production, or clears a delivery requirement that Sora 2 cannot. If 720p Sora 2 passes the craft bar, more volume strengthens the case for staying on the $0.10 rate.

Both Sora 2 and Sora 2 Pro take text or images as input and return video with synced audio. The public pages label both models Legacy. Pro is not a bulk-purchase plan. It is a different model with more expensive resolution choices.

The published 1080p price outruns the public schema

The 1080p option needs verification before it enters a production estimate. OpenAI's live Sora 2 Pro page lists 1080x1920 and 1920x1080 at $0.70 per second. The live Videos API VideoSize schema lists only 720x1280, 1280x720, 1024x1792, and 1792x1024.

Those two official surfaces do not line up. The safe response is not to declare 1080p impossible; it is to test the accepted size in your own project before quoting a client, then keep a 1024p or alternative-model fallback in the delivery plan.

Rate limits can change the production schedule

The free API tier is not supported for either model. On Sora 2, OpenAI lists 25 requests per minute at Tier 1, then 50, 125, 200, and 375 at Tiers 2 through 5. On Sora 2 Pro, the limits are lower: 10, 25, 50, 75, and 150 requests per minute across Tiers 1 through 5.

RPM means requests per minute. It controls how many generation jobs your project can submit, not how much each second costs. A high-volume content pipeline can therefore have enough cash for a render batch and still hit a submission ceiling. Pro's Tier 1 cap of 10 RPM is the first operational limit to model when many variations need to start together.

OpenAI says usage tiers rise automatically as API spend increases. That can improve throughput, but it does not create a pricing break-even. Higher rate limits let you spend the same per-second rate faster.

Is Sora free or included with ChatGPT?

Sora is not free and is no longer included with a current ChatGPT plan. The consumer web and app experiences ended on April 26, and both Sora API model pages say the API free tier is not supported.

The confusion comes from a historical offer that was once accurate. At the 2024 consumer launch, OpenAI included Sora with ChatGPT Plus at no additional cost, with up to 50 videos at 480p or fewer videos at 720p each month. ChatGPT Pro included 10 times more usage, higher resolutions, and longer durations. The old app could reach 1080p and 20 seconds.

None of those consumer allowances is current. Paying for ChatGPT Plus or Pro now does not reopen Sora, and there is no live Sora plan for a student discount to reduce.

The person who never needs to pay is the person who does not need another generation. If your only remaining job is preserving old work, follow OpenAI's export path and move the files into your own storage. Anyone creating new Sora video through the remaining API needs paid API access.

This split matters for procurement. A subscription receipt from the old app and an API usage bill answer different questions. The old bundle priced access by plan. The remaining API prices output by second. Treating the two as interchangeable hides both the real usage cost and the shutdown deadline.

The shutdown is the largest hidden cost

The expensive part of Sora is not the $0.10 starting rate. It is replacing a workflow that stops accepting API requests on September 24.

OpenAI tells users to export Sora content as soon as possible. After discontinuation and any final export window, the company says it will permanently delete data associated with Sora use. The export route is sora.chatgpt.com/sunset, where an export request is delivered by email when ready.

For a creative team, the replacement work includes prompt migration, reference-asset handling, output validation, safety review, storage paths, cost monitoring, and any automation downstream of the generated file. A $0.02 per-second saving disappears quickly if one engineer and one editor have to rebuild that chain under a deadline.

Old credit balances need their own decision. OpenAI's shutdown notice says purchased ChatGPT/Sora credits can still be used for Codex. Its current flexible-credit policy says credits are non-refundable except where law requires otherwise, remain valid for 12 months from purchase, and do not roll over after expiry. Credit salvage is useful, but it does not extend Sora's API life.

Refunds are requests, not an automatic consequence of the shutdown. OpenAI routes Sora refund questions to its ChatGPT subscription process. For direct web purchases, accidental purchases are generally eligible when the company is contacted within 14 days, and eligible refunds are processed within 5 to 7 business days. Apple handles Apple subscription refunds directly. Eligible Google Play refunds are processed within 10 business days, with the same general 14-day accidental-purchase window.

The broader Sora shutdown and alternatives guide explains the product closure. For a larger buyer shortlist beyond the two cost-normalized options below, use the AI video generator comparison.

Sora versus Runway and Google Veo on normalized cost

Sora's $0.10 rate is competitive, but price parity cannot cancel an end date. Runway and Google Veo deserve the budget comparison because both expose current first-party API rates that can be normalized to generated time.

This is a cost comparison, not a claim that the models produce equal quality. Keeper rate, prompt adherence, product geometry, text artifacts, and editability can reverse the cheapest rate-card result.

Runway: a 20% rate premium with a separate API wallet

Runway prices Gen-4.5 at 12 API credits per generated second, and each API credit costs $0.01. That makes the current rate $0.12 per second, or $7.20 per generated minute.

Runway API pricing page showing Gen-4.5 credit cost per generated second
Runway API Pricing

The API requires a minimum $10 credit purchase to start. Runway also warns that credits and plans bought in its web app are completely separate from API credits. A creator with an existing Runway subscription cannot assume that balance funds an integration.

Runway costs $1.20 more than Sora 2 for each generated minute. At 960 generated seconds, the four-attempt campaign scenario costs $115.20, compared with $96 on Sora 2. The $19.20 difference is modest beside a forced API migration, but Runway still has to earn the choice through a keeper-rate test.

The wall is billing fragmentation. App exploration and API production use separate wallets, so a team can pay twice while assuming it has one pool. Put the API organization, minimum credit purchase, and app subscription on separate budget lines.

Google Veo: price parity at Fast 720p and a cheaper Lite path

Google Veo 3.1 Fast with audio costs $0.10 per second at 720p, exactly matching Sora 2, and $0.12 at 1080p. Veo 3.1 Lite with audio costs $0.05 at 720p and $0.08 at 1080p. Neither has a free API tier.

Google Gemini API pricing page showing Veo 3.1 Standard, Fast, and Lite per-second rates
Google Veo 3.1 API Pricing

At 720p, one generated minute costs $6 on Veo Fast or $3 on Veo Lite. Veo Standard with audio costs $0.40 per second at 720p or 1080p, so its generated minute costs $24. Standard is a fidelity tier, not the default budget answer for high-volume concept work.

Veo's current duration choices are 4, 6, or 8 seconds. At 1080p and 4K, the duration must be 8 seconds; Veo Lite does not support 4K. That duration rule affects edit planning even when the per-second rate is clear.

The cost decision is sharp. Veo Fast 720p matches Sora 2's rate without asking a creative operator to build around Sora's September deadline. Veo Lite halves the rate for high-volume iteration. Veo Standard costs four times Sora 2, so it needs a substantially better keeper rate or a final-delivery requirement to pay back.

The same brief must decide the model, not the rate card

A price comparison becomes useful only when every model receives the same production constraint. Use a concrete eight-second DTC skincare brief: one approved serum bottle, one slow camera move, one surface, one lighting direction, no generated on-pack copy, and one clean area reserved for typography in post.

The loose first pass would ask for a “cinematic luxury serum ad” and leave the package, camera, lettering, and edit point open. That prompt invites four expensive variables at once. A better production brief supplies the approved pack as an image reference, fixes a single camera move, bans generated label text, names the eight-second delivery target, and moves the headline into the editor where typography can be controlled.

That is the before-to-after improvement to measure. Score continuity, product geometry, unwanted text, clean edit points, and the number of attempts before one clip clears review. A model that costs twice as much per second but needs half as many attempts ties on keeper cost. A cheaper model that creates more repair work loses despite the rate card.

For the 30-clip campaign at a four-attempt assumption, the cost-only baseline is $96 for Sora 2, $115.20 for Runway Gen-4.5, and $96 for Veo Fast 720p, before vendor duration granularity creates waste. Sora 2 Pro's published 1080p rate reaches $672. Those are budget hypotheses to test, not output-quality results.

Who should still pay for Sora?

Only an existing API user with a short, bounded job should keep paying for Sora. The work must finish before September 24, the team must export what it needs, and the integration must have no continuing dependency after shutdown.

Use Sora 2 at $0.10 per second when 720p clears the delivery bar. Use Pro only when its model or higher-resolution option improves keeper rate enough to beat a 3x, 5x, or 7x price multiplier. Verify 1080p acceptance in the live API before promising it.

Skip Sora for a new product feature, recurring ad program, customer-facing generator, or any workflow expected to run in the fourth quarter of 2026. The cheapest migration is the one completed before more prompts, assets, and automation attach themselves to the ending API.

Decision flow for finishing a short Sora job or migrating an ongoing video workflow
Existing work that ends before September 24 can finish and export; every continuing workflow should migrate now.

The Monday move

The next week should produce a migration decision and a keeper-cost number, not another abstract tool comparison.

  1. Export before testing anything

    Request the Sora archive from the sunset export page. Copy prompts, source images, final outputs, and the metadata your edit or automation depends on into storage your team controls.

  2. Freeze one production brief

    Choose an eight-second delivery your team already makes. Lock the reference asset, aspect ratio, target resolution, camera move, and review criteria. Remove generated typography from the brief and add it in post.

  3. Buy four attempts, not one demo

    Run four attempts on the remaining Sora integration and on the alternatives you can support. Match 720p and delivered duration as closely as each API allows. Record generated seconds, accepted clips, repair time, and every billing minimum.

  4. Calculate keeper cost

    Divide total generation spend by accepted clips, then add the editor and engineering time needed to repair or route each output. The model with the lowest rate is not the winner unless it also produces the lowest acceptable-outcome cost.

  5. Move the production path

    Select the replacement before Friday, stop new Sora-dependent work, and set an internal cutoff ahead of September 24. Keep Sora only as a temporary source during the handoff, never as the destination architecture.

That is the practical consequence of Sora pricing in 2026. The rate card still looks usable. The product timeline says the budget has already moved.

Sora pricing FAQ

How much does Sora cost?

There is no current Sora consumer subscription. Until September 24, 2026, Sora 2 API access costs $0.10 per generated second at 720p, while Sora 2 Pro costs $0.30 at 720p, $0.50 at 1024p, and a published $0.70 at 1080p.

How much does an eight-second Sora video cost?

Sora 2 costs $0.80. Sora 2 Pro costs $2.40 at 720p, $4.00 at 1024p, and a rate-card $5.60 at 1080p. The public Videos API size schema does not currently list a 1080p value, so verify that option before quoting it.

Is Sora free with ChatGPT?

No. Sora was historically included with ChatGPT Plus and Pro, but the web and app experiences closed on April 26, 2026. The remaining Sora 2 and Sora 2 Pro APIs do not support the free usage tier.

Does Sora have a student discount?

No current Sora subscription exists to discount. Students face the same API rate card, and the API free tier is not supported.

What is Sora's refund policy?

OpenAI's shutdown page sends subscription refund questions to the ChatGPT refund process and says purchased ChatGPT/Sora credits can still be used for Codex. Accidental direct-web purchases are generally eligible when reported within 14 days; Apple handles Apple subscription refunds directly.

Do Sora credits expire or roll over?

OpenAI's current flexible-credit policy says credits remain valid for 12 months from purchase, then expire without rollover, and are non-refundable except where required by law. The Sora shutdown notice says purchased ChatGPT/Sora credits can still be used for Codex.

Did Sora pricing change in 2026?

Yes. Consumer access through the Sora web app, mobile app, and historical ChatGPT bundles ended on April 26. Only usage-priced API access remains, and that ends on September 24.

Does the Sora API have a 50% Batch discount?

Do not budget one. OpenAI's current Sora model pages publish only the standard $0.10, $0.30, $0.50, and $0.70 per-second rates, with no Sora-specific Batch price.

Why is OpenAI shutting down Sora?

OpenAI's current public discontinuation notice gives the April and September closure dates, export instructions, deletion warning, and refund route, but it does not state a reason for the shutdown.

Use the AI Business Workflow Audit Checklist to decide which video workflow deserves a continuing budget line.

Last Updated

Aug 13, 2026

CategoryDesign

More from Design

View all Design articles
Newsletter

One letter, every Sunday. Working systems, not hot takes.

Build logs, working systems, and field notes from running a portfolio of AI ventures.

Weekly. No spam. Unsubscribe anytime.