Sora alternatives after the API shutdown

What is shutting down, and when

The Sora wind-down happened in two stages, and only the second one is still ahead of you.

Two details matter for planning. First, the deprecation notice was published on March 24, 2026, which means the six-month window is nearly spent. Second, OpenAI lists no recommended replacement — unlike a normal model retirement where you swap a model string, this one requires moving to another vendor's API and rewriting the request and job-polling code around it.

Added 2026-09-15: leaving does not cost you anything on the meter. This page originally ranked the destinations without pricing the thing you are leaving, because Sora's own rates were not published here. They are now, on the Sora public-data review, and they change the tone of the migration: OpenAI's model reference lists sora-2 at $0.10 a second at 720p and sora-2-pro at $0.30, $0.50 and $0.70 a second by resolution, against Veo 3.1 Fast's identical $0.10 at 720p and Veo 3.1 Lite's $0.05. Matched resolution for resolution, every exit from Sora is cheaper or the same — from break-even at the bottom to −82.9% moving a 1080p Pro workload to Veo 3.1 Fast. The cost of this migration is engineering time, not spend.

Updated 2026-09-21: check your clip-length assumptions before you port the code. Every Veo and Kling rate above was re-verified today and none has moved. What did change is the Sora side: the Sora profile previously reported a hard twelve-second ceiling taken from OpenAI's Videos API reference, but OpenAI's video generation guide and its own prompting cookbook both document 20-second generations and extension to 120 seconds total, and two sources beat one. That matters for a migration rather than for a budget: if you are porting a pipeline that stitched short Sora segments, the destinations cap a single generation shorter than Sora documented — Kling at 15 seconds, Veo at 8 per generation in Flow, Seedance 2.5 at 30 in one pass — so the stitching logic you were hoping to delete is likely to be the part you keep.

Correction 2026-09-23: there was a batch tier, and it halves the saving. The 2026-09-15 note above priced the migration off Sora's standard rates only. OpenAI's API pricing page also carries a batch column for both video models at exactly 50% of the standard rate on all four configurations — sora-2 at 720p for $0.05 a second, and sora-2-pro at $0.15, $0.25 and $0.35 — and CartSignal had not published it. The column is absent from the sora-2 and sora-2-pro model reference pages, which is where the earlier rates were read. Two things change. The claim that every matched-resolution exit is cheaper or break-even still holds measured from the standard rates, but measured from the batch rates those margins roughly halve, and from Sora's batch floor of $0.05 the best available exit is a tie with Veo 3.1 Lite rather than a saving. And if the pipeline you are porting ran on the Batch API, budget the destination against $0.05 rather than $0.10, because that is the number your invoices actually carried. The full rate card is on the Sora profile; the field-wide comparison is on what AI video generation costs per second.

The shortlist at a glance

Google Veo 3.1

Veo is the most direct substitute if what you valued about Sora was a single text prompt producing a finished clip with sound. There are two ways in, and they are priced completely differently.

Through the API. Google's official Gemini API pricing page lists Veo 3.1 in three variants, all charged per second of output with audio included by default: Standard at $0.40/second for 720p and 1080p and $0.60/second for 4K; Fast at $0.10/second (720p), $0.12/second (1080p) and $0.30/second (4K); and Lite at $0.05/second (720p) and $0.08/second (1080p), with no 4K option. The page also notes you are only charged when a video is generated successfully. At Fast 1080p rates, an eight-second clip works out to roughly a dollar — useful for sizing a migration budget before you write any code.

Through Flow. If you are a creator rather than a developer, Google's Flow studio runs Veo 3.1 on a credit allowance tied to your Google AI subscription. Google's Flow credits help page lists a free tier with 50 daily credits, then Google AI Plus at 200 monthly credits, Google AI Pro at 1,000 credits, and the two Google AI Ultra tiers — labelled $100 and $200 by Google — at 10,000 and 25,000 credits. Credits do not roll over, and Google notes features vary by tier, platform and region, so confirm what your country actually gets. Full breakdown of both pricing routes: Google Veo public-data review.

One update since this page first ran: Veo is no longer Google's only video model. Gemini Omni Flash, announced at Google I/O on May 19, 2026, is a separate multimodal model that also outputs video at roughly $0.10 per second, and both it and Veo 3.1 remain listed on the Gemini API pricing page. It is worth pricing alongside Veo before you commit a migration — details on the Veo profile.

Kling

Kling is the alternative most often named alongside Veo in public model comparisons, and its recent versions generate with native audio. It sells a consumer credit subscription rather than per-second API pricing.

Update, 2026-08-31: Kling's membership page still blocks direct retrieval, so its plan prices remain third-party reported rather than verified. But the credit mechanics are documented officially, and they matter more than the sticker price. Kling's official Kling 3.0 model guide bills per second of output: 6 credits/second at 720p and 8 at 1080p without audio, rising to 9 and 12 credits/second with native audio. A five-second 1080p clip with sound is therefore 60 credits — and on the entry plan's reported 660 monthly credits, that is about eleven clips a month before you re-roll a single failed prompt.

Two traps carry over from the Sora situation. Plan credits expire at the end of each billing cycle and do not roll over, and pricing trackers consistently describe the advertised rate as a discounted first month that renews higher. For developers, Kling's API is reported to start from $0.075/second — between Veo Lite's $0.05 and Veo Fast's $0.10–$0.12, so not the decisive saving it first appears. Full breakdown, including cost per finished clip on each tier: Kling AI public-data review.

Runway

Runway is the pragmatic pick if you do not want to bet a workflow on one model. Its official pricing page lists Free ($0, 125 one-time credits), Standard at $12/month billed annually ($15 monthly, 625 credits/month), Pro at $28/month annually ($35 monthly, 2,250 credits), Max at $76/month annually ($95 monthly, 9,500 credits), and custom Enterprise packages, with a 20% annual discount across paid tiers.

The more interesting part is the model list. Runway's pricing page names not only its own Gen-4.5 and Aleph 2.0 but also Seedance, Kling 3.0 and Veo 3.1 — so a single Runway subscription can be a hedge against exactly the situation Sora users are in now. Confirm which models your specific tier includes, since access to third-party models is the sort of thing that moves between plans, and note that Kling appears in Runway's app lineup but not on its developer API pricing page.

Two things changed in September 2026 that affect this recommendation. First, Runway retired the Unlimited plan: its help centre states that existing Unlimited subscriptions switched automatically to the credit-metered Max plan on 1 September 2026, and Explore Mode — which generated without spending credits — did not carry over. Every tier is now capped. Second, the consolidation saving is real only at the top: Runway charges 40 credits per second for Veo 3.1 with audio, which works out to $0.32/second on Max billed annually against Google's own $0.40/second list price, but $0.96/second on Standard billed monthly. Routing Veo through Runway on a cheap plan costs more than going direct to Google. Full breakdown of the credit rates, what each tier buys per month, and the cost of a finished clip: Runway public-data review.

If Sora was not really the right tool anyway

A lot of people reaching for Sora were trying to make something a prompt-to-video model is bad at. If your actual output is a person talking to camera, look at HeyGen or Synthesia instead — the HeyGen vs Synthesia comparison covers that split. If it is faceless social video built from a script or an article, Fliki and Pictory are built for that pipeline. If you mostly needed to cut and caption footage you already have, Descript and VEED are editors rather than generators, and cost far less per finished minute.

Migration checklist

• Export anything still on OpenAI infrastructure now. The API date is a removal date, not a read-only date.
• Budget in seconds, not clips. Veo bills per second of output, so a per-clip cost from your Sora usage will not translate directly.
• Re-test your prompts. Prompt phrasing does not transfer cleanly between video models; expect to rewrite your prompt library, not port it.
• Check audio separately. Veo and recent Kling versions generate audio natively, which changes whether you still need a separate voice tool from the AI voice category.
• Read the rights terms before republishing. Commercial-use, likeness and watermarking rules differ by vendor and by plan tier.
• Avoid single-vendor exposure. Sora is the clearest recent case of a well-funded video model being switched off; a platform that hosts several models, or a thin abstraction of your own, limits the damage next time.

FAQ

When exactly does the Sora API stop working?

OpenAI's public deprecations page lists the Videos API and the sora-2 and sora-2-pro model aliases and snapshots for removal from the API on September 24, 2026. The notice was published on March 24, 2026 and names no replacement model. Re-verified 21 September 2026, the date has not moved — and OpenAI does not document what a request made after that date returns, so treat the response as absent rather than as a specific error you can catch.

Is there a drop-in replacement for the Sora API?

No. OpenAI does not list a successor video model, so migrating means moving to a different vendor's API, such as Google's Veo models on the Gemini API, and rewriting your request and polling code against that vendor's schema.

What is the closest alternative to Sora for quality?

Google's Veo 3.1 and Kling are the two most frequently named alternatives in public model comparisons, and both generate video with native audio. CartSignal has not run hands-on quality tests, so treat any ranking as public positioning rather than a measured result.

Can I still download videos I made in Sora?

It depends which half of Sora you used, and the API answer surprises people. For API output there was never a durable library to download from: OpenAI's video generation guide states that download URLs are valid for a maximum of 1 hour after generation and instructs callers to copy files to their own storage promptly, with batch output available for up to 24 hours. So if your pipeline saved each MP4 at generation time, the shutdown takes nothing away and you lose only the ability to generate more; if it kept OpenAI URLs and re-fetched them, those were already expiring hourly. For the consumer app, discontinued on April 26, 2026, OpenAI directed users to an export flow before shutdown; third parties report that flow covers app and web generations rather than API ones, but CartSignal could not verify it, because help.openai.com returns HTTP 403 to our fetches.

Do I need a separate subscription for each video model?

Not necessarily. Runway's pricing page lists several third-party models, including Veo 3.1 and Kling 3.0, alongside its own Gen-4.5, so one credit-based subscription can cover more than one model. Everything is spent from a single credit pool at published per-second rates, so check the burn rate as well as the model list: Veo 3.1 with audio costs 40 credits per second through Runway, which only beats Google's own $0.40/second list price on the Max plan billed annually. Kling also appears in Runway's app lineup but not on its developer API pricing page.