Sora Review (Verified August 2026)
Sora is closed to consumers and its API ends September 24. See current prices, capabilities, hard limits, and the best active alternatives.
Sora is not worth adopting in August 2026: the consumer product is already closed, and its remaining API shuts down in 39 days. Existing API teams may finish a bounded job, export everything, and migrate; every new buyer should choose an active alternative.
What Sora is in August 2026
Sora is a deprecated video-generation API, not a consumer creative tool you can join. OpenAI closed the Sora web and app experiences on April 26, 2026, and will close the remaining API on September 24, 2026. Until then, existing API projects can send text or image inputs to Sora 2 or Sora 2 Pro and receive short video with synced audio. The old social feed, consumer interface, and ChatGPT subscription access are gone. OpenAI’s former Sora homepage now leads visitors to the discontinuation notice, which is the clearest description of the product’s current state. I verified that status, every rate below, and the current developer documentation on August 16, 2026. The model can still render; the product cannot support a new long-term workflow. OpenAI’s live shutdown notice is the source of truth.
Who Sora is for, and who should skip it
Sora is for one narrow group: existing API users whose final job, export, and migration can all finish before September 24. It is a skip for a creator comparing subscriptions, an agency starting a recurring client workflow, or a product team choosing a video backend.
The shortlist is unusually easy because lifecycle overrides small differences in output quality. An active platform can improve, fix bugs, preserve integrations, and support next quarter’s work. Sora cannot promise any of those things beyond its published end date.
Stay with Sora only when all four conditions are true: the integration already exists, the job is bounded, the client accepts the available output, and the production archive will be outside OpenAI before September 24. Missing any one condition changes the answer to migrate now.
Choose Runway for an active production platform
Runway is the best skip-to option for a creator who wants current Gen-4.5 access, no-watermark exports on paid plans, and 4K upscaling in the same account. Its Free tier includes 125 one-time credits. Standard costs $15 monthly, or $12 per month on annual billing, and includes 625 monthly credits, listed as 52 seconds of Gen-4.5. Pro costs $35 monthly, or $28 annually, for 2,250 credits and 187 seconds. Max costs $95 monthly, or $76 annually, with 9,500 credits and one month of rollover. Those allowances are not generous if every prompt needs repeated takes, but Runway is at least a platform you can still buy and plan around. Runway publishes the live plan details here.
Choose Runway when the job needs an active model catalog and a predictable paid workspace more than it needs Sora’s lowest $0.10 per-second rate. Skip it when 52 top-model seconds on Standard would disappear during concepting alone.
Choose Google Flow for Veo and scene construction
Google Flow is the strongest no-cost starting point for a creator who wants to build shots into scenes rather than buy a dying API. Its free access includes 50 daily Flow credits plus text-to-video, frames-to-video, ingredients-to-video, extensions, video-to-video editing, Scenebuilder, characters, and avatars. Google AI Plus is $4.99 per month with 200 monthly credits and 1080p upscaling. Pro is $19.99 with 1,000 credits. The two Ultra levels are $99.99 with 10,000 credits and 4K upscaling, or $199.99 with 25,000 credits. Google lists the current Flow access and tiers on the product page.
Choose Flow when a scene-based Veo workflow and a usable free allowance matter. Skip the higher plans until the same brief has proved that the additional credits and upscaling change the deliverable, not just the invoice.
Choose Adobe Firefly for an Adobe-centered workflow
Adobe Firefly is the practical choice for a team already moving images, video, audio, and partner-model outputs through Adobe tools. Free users receive daily generations. Standard costs $9.99 per month with 2,000 generative credits and up to 20 five-second videos. Pro costs $19.99 with 4,000 credits and up to 40. Pro Plus regularly costs $49.99 with 10,000 credits and up to 100, while Premium regularly costs $199.99 with 50,000 credits and unlimited access to Adobe’s Firefly Video Model. Adobe is temporarily advertising Pro Plus at $34.97 and Premium at $139.91 through August 26, 2026. Adobe’s plan page carries the current allowances and promotion.
Choose Firefly when the workflow already ends in Adobe and a mixed image, video, and audio credit pool is useful. Skip Premium until volume justifies the regular $199.99 price, because a temporary discount should not define a permanent production budget.
The broader Sora shutdown and alternatives guide covers more replacement options. For this review, the decision is simpler: continuing to evaluate a product after its exit date has been published is not optionality. It is delayed migration.
Capability 1: Prompt or reference frame to synced video
Sora’s strongest remaining capability is turning a text prompt or a matching reference frame into a short video with audio through one API job. That is useful when a team already has software around the endpoint and needs a bounded set of shots before shutdown.
The craft improvement comes from controlling variables before paying for a render. Take an eight-second landscape product-film brief for a skincare bottle. A loose first pass such as “cinematic luxury serum ad” leaves the package geometry, camera, surface, lighting, action, and edit point open. Even an attractive result can be unusable if the bottle turns, the label mutates, or the camera move leaves no clean frame for type.
The controlled version starts from the approved bottle image at 1280x720, because OpenAI requires the image reference to match the requested output resolution. The prompt then names the five ingredients OpenAI recommends: shot type, subject, action, setting, and lighting. One usable direction would be: a locked medium product shot of the approved bottle, one slow clockwise camera arc, bottle stationary on pale stone, hard morning side light, clean negative space at frame left. Brand copy belongs in the editor, not inside the generated plate.
Fix the deliverable first
Choose one eight-second 1280x720 shot, one camera move, and one clean edit point. Eight seconds appears in the current API reference and avoids building the test around a disputed duration.
Prepare the first frame
Export the approved reference at exactly 1280x720. The reference becomes the opening frame, so composition and empty space should already serve the final layout.
Write the shot, not the mood
Specify shot type, subject, action, setting, and lighting. Keep the action singular. Add typography and legal copy later in the edit where spelling and placement stay controllable.
Draft on Sora 2, finish only if earned
Sora 2 costs $0.10 per second and is the cheaper iteration tier. Move a selected shot to Sora 2 Pro only when the higher model or resolution clears a visible delivery requirement.
Download and archive immediately
The completed MP4 download URL remains valid for a maximum of one hour. Copy the file, prompt, reference, settings, and review result into storage the team controls.
This workflow is ship-ready only as a source-plate process. Sora can create motion, framing, and synced sound. The final piece still needs controlled typography, color review, audio decisions, captions, and an edit. Treating one generated clip as a finished advertisement lowers the craft bar exactly where a professional workflow should raise it.
Capability 2: Reusable non-human characters
Sora can preserve a mascot or object across multiple shots through reusable character assets, but the current API makes this a non-human continuity tool by default. Human-likeness uploads are blocked unless an account has special eligibility.
The useful workflow is a brand mascot, product, or creature that needs the same silhouette and surface treatment in several clips. OpenAI recommends a short 2-to-4-second source video in 16:9 or 9:16 at 720p to 1080p. A single generation can contain up to two characters. The character’s name must also appear verbatim in the prompt; sending the asset ID alone is not enough for reliable preservation.
For a coffee brand, that could mean creating a reusable ceramic-cup mascot from a short turntable clip, naming it “Mossy,” and then generating a sequence of discrete eight-second scenes. The first brief establishes the mascot walking through a café. The second keeps the same name and asset while changing only the action and camera. A reference frame can also guide the opening composition of an individual generation.
The wall appears when the sequence needs to continue. Extensions do not support characters or image references. A team can either use the reusable character in separate shots or use extension context to continue motion, but it cannot rely on both controls inside the extension. That incompatibility is more important than the friendly “characters” label because it decides whether a storyboard can survive beyond isolated clips.
Sora 2’s system card credits the model with improved physics, realism, steerability, and synced audio. Those are meaningful model qualities. They do not remove the workflow constraint: a consistent subject must remain controllable across the exact operations a production uses, not merely inside one successful shot.
Capability 3: Focused edits, extensions, and batch queues
Sora supports a sensible finishing loop through targeted edits, video extensions, and batch creation, but each operation has a different compatibility boundary. A production team must design around those boundaries before it sends a shot list.
A focused edit is the cleanest operation. OpenAI recommends changing one thing at a time because a narrow instruction preserves more of the source structure and composition. If the approved clip needs a teal, sand, and rust palette, ask for that palette change alone. Do not combine color, camera, character, and location changes in one request and expect the accepted parts of the shot to remain fixed. Editing an uploaded external video is restricted to eligible customers, while editing an existing Sora video can use its video ID.
Extensions continue a completed clip while using the full source as context. Each extension can add up to 20 seconds, and a video can be extended up to six times for a 120-second maximum. That sounds generous until the control tradeoff lands: an extension accepts the source video and prompt, but not a character asset or image reference. Use it for a continuing camera move or environmental reveal, not for a character-dependent brand sequence.
Batch is the right shape for a preplanned shot list. Each line carries one create-video request, so a studio can queue many independent prompts and map the results back to internal shot IDs. Batch currently supports creation only, uses JSON instead of multipart uploads, and requires reference assets to be uploaded first. Results remain downloadable for up to 24 hours after the batch completes.
A bounded campaign workflow would look like this:
- Lock four independent eight-second shots and upload every approved reference in advance.
- Queue four prompt variants per shot through Batch, with stable IDs such as
hero-01-aandhero-01-brecorded in the production tracker. - Review completed clips against geometry, motion, audio, clean edit points, and brand-safe negative space.
- Apply one focused edit only to a clip that is already close to approval.
- Extend the final environment shot only if losing character and image-reference inputs is acceptable.
- Download every asset before the 24-hour batch window closes, then remove Sora from the continuing path.
This is the point where Sora still looks competent. The API has enough surface for an orderly render pipeline. The problem is that every hour spent polishing that pipeline produces code and process that must be replaced within 39 days.
Sora pricing: every current tier
Sora has no current consumer price, free plan, monthly tier, or annual tier. The only live generation prices are the four API rates below, and the API free tier is not supported.
Every number was verified against OpenAI’s live model pages on August 16, 2026. The dedicated Sora pricing analysis covers rate limits, per-minute normalization, and the separate 1080p schema issue in more detail.
The useful budget is not one lucky render. A 30-second spot needs at least four eight-second source shots, or 32 seconds before trimming. Assume four candidate renders for each of those four shots. That is 16 renders and 128 generated seconds. This is a planning model, not a claimed Sora acceptance rate.
- Sora 2 at 720p costs $12.80 for that candidate set.
- Sora 2 Pro at 720p costs $38.40.
- Sora 2 Pro at 1024p costs $64.00.
- Sora 2 Pro at the published 1080p rate costs $89.60.
The model choice has no volume break-even because every tier remains linear per second. Pro is three times the base rate at 720p, five times at 1024p, and seven times at the published 1080p rate. Pro pays back only when its output clears a resolution or quality requirement that base Sora 2 cannot, or when it reduces enough failed attempts and repair time to offset the multiplier.
A single 20-second 1080p Pro generation would cost $14 at the published rate. That figure also exposes why the documentation conflict matters: a client estimate should not assume a duration and size simply because one official guide advertises it when another official API surface omits it.
The limitations that decide the review
Sora’s largest limitation is not a warped frame or a strict prompt filter. It is the published end of the product, followed by documentation and workflow constraints that make even the remaining 39 days harder to budget.
1. The API disappears on September 24
No prompt technique fixes lifecycle risk. The shutdown affects Sora 2, Sora 2 Pro, and the named snapshots in the video guide. A customer-facing generator, recurring ad system, or studio tool that depends on those models needs a replacement. OpenAI also tells former consumer users to export promptly and says associated data will be permanently deleted after discontinuation and any final export window.
This makes Sora unsuitable for procurement, onboarding, or a new integration even if one clip looks better than an alternative. A production tool is the contract around the output as much as the output itself.
2. OpenAI’s current duration and size pages disagree
The live video-generation guide says both models support 16-second and 20-second generations and tells users to choose Sora 2 Pro for 1920x1080 or 1080x1920 output. The live API reference lists only 4, 8, and 12 seconds, with size values topping out at 1024x1792 or 1792x1024.
That conflict is the named limitation most competing reviews miss. It does not prove the longer or 1080p jobs are unavailable to every project. It does mean a producer should run an accepted-request check in the exact account before quoting a 20-second or 1080p delivery. Budget from the public schema until the endpoint confirms otherwise.
3. Human-led brand work hits hard guardrails
The current API guide says only content suitable for audiences under 18 is allowed, copyrighted characters and music are rejected, real people cannot be generated, and input images with human faces are rejected. Human-likeness character uploads are blocked by default and require account-level eligibility.
These controls are defensible safety choices, but they disqualify ordinary creative briefs. A fashion campaign built around an approved model reference, a founder-led launch video, or a licensed character concept cannot assume the standard API path will accept its source material. Use a non-human mascot, change the concept, or choose a platform whose approved workflow matches the brief.
OpenAI also says Sora videos carry provenance signals, including visible watermarks at launch and C2PA metadata. C2PA is embedded provenance information that helps software identify where media came from. That is useful for disclosure and may still require a separate finishing decision for the final asset.
4. Continuity controls do not travel across operations
Sora characters can aid consistency, but extensions do not accept characters or image references. Image references must match the target resolution. Uploaded-video editing is restricted to eligible customers. Batch creates videos but does not batch the extension and edit stages.
Each limit is manageable in isolation. Together they force a studio to choose which control it is willing to lose at each step. The workflow is strongest for discrete shots with one approved opening frame and one action, weaker for a long sequence whose subject, framing, and motion must survive every operation.
5. Render and asset timing create operational work
OpenAI says one render may take several minutes depending on the model, load, and resolution. A direct completed-video download URL lasts at most one hour, while a batch result remains available for up to 24 hours. Sora 2 starts at 25 requests per minute for API Tier 1, while Sora 2 Pro starts at 10.
That is enough for a queue, not a reason to leave assets sitting on the vendor. The application needs webhooks or polling, failure handling, its own object storage, review state, and cleanup. Building that plumbing for a new Sora integration now would be engineering work with an expiration date.
6. Sora is a plate generator, not a finished design system
Sora can produce motion, sound, and scene continuity. A ship-ready campaign still needs controlled type, brand review, color finishing, legal copy, audio mix, captions, and exact export specs. The best professional use is a generated source plate inside a broader edit. Mood boards and visual exploration can tolerate variation; a locked client deliverable cannot.
This distinction matters because the retry cost is only the visible compute line. Review time, rejected geometry, manual repair, and migration effort can dwarf a $0.80 draft. A tool earns production status when the whole path is repeatable, not when its best sample is impressive.
Sora verdict: finish, export, and leave
Sora is a skip for every new buyer and a temporary bridge for a small group of existing API users. The model’s synced audio, image guidance, characters, edits, extensions, and batch jobs remain capable enough to finish a bounded project. None creates a rational path past September 24.
- Sora 2 starts at $0.10 per generated second for inexpensive 720p drafts.
- Text and matching image references can produce video with synced audio in one job.
- Focused edits, extensions, reusable non-human characters, and Batch cover several useful production patterns.
- Sora 2 Pro publishes 720p, 1024p, and 1080p price levels for higher-finish work.
- The consumer product is closed and the API shuts down September 24, 2026.
- Current guide and reference pages disagree on supported durations and 1080p sizes.
- Human faces, real people, copyrighted characters, and copyrighted music hit strict default restrictions.
- Characters and image references do not carry into extensions.
- Render latency and short download windows require infrastructure that has no future on Sora.
The Monday move is a five-part exit, not another model demo.
Export first
Request the Sora archive from sora.chatgpt.com/sunset and copy every prompt, source, output, and piece of metadata the workflow needs into controlled storage.
Freeze one representative brief
Use one eight-second shot with a fixed reference, action, aspect ratio, resolution, camera move, and review rubric. Do not compare models on different creative problems.
Price a candidate set
Budget four candidates, record generated seconds and review time, and calculate cost per accepted plate rather than price per successful request.
Run the same brief on the replacement
Choose Runway for an active multi-model workspace, Flow for Veo and scene tools, or Firefly for an Adobe-centered pipeline. Judge the accepted output and the work around it.
Set an internal cutoff
Stop new Sora-dependent work before the vendor deadline. Keep the remaining API only as a temporary source while the replacement path becomes the default.
Sora review FAQ
Is Sora AI any good?
Sora 2 remains capable of short video with synced audio, image guidance, focused edits, reusable non-human characters, and extensions. It is still a poor buying decision because the consumer product is gone and the API closes on September 24, 2026.
Why is Sora AI shutting down?
OpenAI’s current public discontinuation notice gives the April 26 consumer closure, the September 24 API closure, export instructions, data-deletion warning, and refund route. It does not state a reason for the shutdown.
Does Sora AI actually work?
The remaining API is live, documented, and priced as of August 16, 2026. Existing users can create and manage video jobs, but OpenAI labels the models and API deprecated, and its public pages disagree on some available durations and sizes.
Is there any AI better than Sora?
Runway, Google Flow, and Adobe Firefly are better purchases because they are active platforms. Pick Runway for Gen-4.5 and 4K finishing, Flow for Veo and scene-building tools, or Firefly when the workflow already lives in Adobe.
What is the problem with Sora AI?
The decisive problem is the shutdown. Additional limits include conflicting duration and resolution documentation, strict default restrictions on human references and copyrighted material, extension incompatibilities, render latency, and short asset-download windows.
How much does Sora cost now?
There is no active consumer tier. Sora 2 costs $0.10 per generated second at 720p. Sora 2 Pro costs $0.30 at 720p, $0.50 at 1024p, and a published $0.70 at 1080p until the API closes.
Can ChatGPT Plus or Pro users still access Sora?
No current ChatGPT consumer plan reopens the discontinued Sora web or app experience. Purchased ChatGPT/Sora credits can still be used for Codex, according to OpenAI’s shutdown notice.
What should existing Sora users do?
Export the archive immediately, finish only jobs that end before September 24, run one representative brief on an active replacement, and set an internal migration cutoff ahead of OpenAI’s deadline.
Use the AI Business Workflow Audit Checklist to decide which creative workflow deserves a continuing budget line.
Aug 16, 2026
