Sora vs Veo: Sora Is Discontinued — What to Use Instead
Sora is discontinued. If you searched for “Sora vs Veo” hoping to pick between the two AI video generators, the choice is already made for you: OpenAI closed the Sora app and website on April 26, 2026, and shut down the Sora API on September 24, 2026, with no replacement named. Veo 3.1, Google’s current video model, is the one you can actually use today. This guide compares them honestly anyway — what Sora was, what Veo is, what Veo costs, and where Sora users should go now — because most of the comparisons ranking on Google still treat Sora as a live product, and that advice is months out of date.
The short answer: pick Veo 3.1. On pure quality the two models were nearly tied — on LMArena’s text-to-video leaderboard, published November 9, 2026, Sora 2 Pro scored 1367 and Veo 3.1 with audio scored 1363, a gap so small that blind voters picked each about as often as the other. But quality is irrelevant when one product no longer exists. Veo 3.1 is available right now through Google Flow, the Gemini app, and Google Cloud, starting at $19.99 a month on Google AI Pro. Sora is not available at any price.
Sora vs Veo at a Glance
| Sora 2 (OpenAI) | Veo 3.1 (Google) | |
|---|---|---|
| Status today | Discontinued — app closed April 26, 2026; API shut down September 24, 2026 | Live and actively developed |
| Max resolution | 1080p | Up to 4K on 8-second clips |
| Clip length | 10 seconds by default, up to 25 seconds with Sora 2 Pro and Storyboard, extendable to about 2 minutes | 4, 6, or 8 seconds per generation; longer sequences are stitched together in Flow |
| Native audio | Yes — speech, sound effects, and ambient sound in the same generation | Yes — dialogue, sound effects, and ambience; a headline feature since Veo 3 |
| Standout feature | Characters — insert yourself or a friend into any scene from one short recording | Ingredients to Video — build a clip from up to three reference images |
| LMArena text-to-video score | Sora 2 Pro: 1367 (#13); Sora 2: 1342 (#19) | Veo 3.1 (Audio): 1363 (#14); Veo 3.1 (Audio, 1080p): 1363 (#15) |
| Price today | No price — nothing to buy | Google AI Pro $19.99/month; API from $0.05/second (Lite) to $0.40/second (Standard, with audio) |
| Watermark | C2PA content credentials embedded in videos | SynthID in every frame and in the audio, plus C2PA credentials signed by Google |
What Happened to Sora?
Sora had one of the shortest lives of any major AI product. OpenAI announced the original Sora in February 2024, calling it the “GPT-1 moment for video,” and released Sora 2 with a social iOS app on September 30, 2025. Less than six months later it was gone. OpenAI announced the discontinuation on March 24, 2026, closed the web and app experiences on April 26, 2026, and removed the Sora 2 models and the Videos API on September 24, 2026. OpenAI’s own help page states both dates plainly, and its API deprecations page lists every Sora 2 model with a shutdown date of September 24, 2026 — and no replacement model named. The Sora 2 launch post itself now carries a notice at the top: “As of April 26, 2026, the Sora product is no longer available.”
| Date | What happened |
|---|---|
| February 2024 | OpenAI announces the original Sora text-to-video model |
| December 2024 | Sora Turbo becomes publicly available |
| September 30, 2025 | Sora 2 launches with a social iOS app and the Characters feature; Sora 2 Pro arrives for ChatGPT Pro users |
| October 6, 2025 | Videos API launches with the Sora 2 and Sora 2 Pro models |
| December 2025 | Disney announces a three-year, $1 billion licensing deal covering 200+ characters for Sora 2 |
| March 24, 2026 | OpenAI announces it is discontinuing Sora in the consumer app and the API |
| April 26, 2026 | Sora web and app experiences close |
| September 24, 2026 | Sora 2 models and the Videos API are removed; no replacement is named |
Why Did OpenAI Shut Down Sora?
OpenAI’s own explanation is short: a spokesperson said the company had decided to discontinue Sora in both the consumer app and the API, tying the move to focus and growing compute demand, while the Sora research team continues its world-simulation work for robotics. The press reporting fills in the arithmetic. The Wall Street Journal, via TechCrunch, reported that running Sora cost roughly $1 million a day, that its user base peaked near 1 million after launch and then fell below 500,000, and that the Disney deal — $1 billion over three years — wound down alongside the product, with Disney learning of the decision less than an hour before it went public. Reuters reported the same compute-strain picture: video generation was consuming resources OpenAI wanted for coding tools and enterprise customers. In other words, Sora was not shut down because the model was bad. It was shut down because a consumer video app burning scarce compute for a shrinking audience was an easy cut once OpenAI decided that compute was worth more elsewhere.
If you made videos with Sora, export them now. OpenAI’s help page points to an export tool at sora.chatgpt.com/sunset: open it, click Export, and you get an email when your content is ready to download. OpenAI says it will permanently delete data associated with your use of Sora after the discontinuation and after any final export window passes, so do not assume your old clips will sit there indefinitely.
What Was Sora 2 Good At?
It is worth remembering what Sora 2 actually did well, because the comparison only makes sense against the real thing. Sora 2 was a genuine leap over the original Sora: OpenAI described it as the “GPT-3.5 moment for video,” and its party trick was physics. Where earlier video models would morph objects to satisfy a prompt — a missed basketball spontaneously teleporting into the hoop — Sora 2 modeled failure correctly, letting the ball rebound off the backboard. It could follow intricate multi-shot instructions while keeping the world state consistent, and it handled realistic, cinematic, and anime styles in the same model. It also generated synchronized audio in the same pass: background soundscapes, speech, and sound effects, not a silent clip with music added later.
Its most talked-about feature was Characters. After a short one-time video-and-audio recording to verify your identity, you could drop yourself — or a friend, a pet, or an object — into any Sora scene with your appearance and voice intact, and revoke access at any time. The social app was built around that idea: a feed of remixes where friends inserted themselves into each other’s generations. And on the numbers, Sora 2 held its own against everything on the market. On LMArena’s text-to-video leaderboard, published November 9, 2026, Sora 2 Pro scored 1367 and standard Sora 2 scored 1342 — effectively tied with Veo 3.1, which scored 1363. One independent comparison also measured Sora 2’s text-rendering readability at 84%, well ahead of Veo 3.1’s 41% on the same test. None of that matters now, because you cannot generate a single frame with it. But it explains why the shutdown surprised people: this was a competitive model, killed by economics, not by quality.
What Is Veo 3.1?
Veo 3.1 is Google DeepMind’s current flagship video model, launched in October 2025 as an update to Veo 3. Its headline feature is native audio: dialogue with lip-sync, sound effects, and ambient sound generated in the same pass as the picture, at 48kHz stereo. It produces 4-, 6-, or 8-second clips at 24 frames per second in 16:9 or 9:16, with 4K output on 8-second clips. Its most useful creative control is Ingredients to Video: hand it up to three reference images — a character, an object, a scene — and it builds a clip around them with consistent characters across generations. It also offers scene extension, first-and-last-frame control, and object insertion for directed shots.
On Google’s own benchmarks, Veo 3.1 leads. DeepMind’s Veo page reports that on MovieGenBench, Meta’s 1,003-prompt benchmark set, Veo 3.1 performed best on overall preference, text alignment, and visual quality, and best on audio-visual preference and audio-video alignment on the 527-prompt audio subset — all Google-reported figures from October 2025, so read them as the vendor’s own scorecard, not an independent test. The independent picture is closer: LMArena’s November 9, 2026 text-to-video board puts Veo 3.1 (Audio) at 1363, essentially tied with Sora 2 Pro’s 1367. The same board also carries a warning for anyone buying into Veo as the final answer: the current #1 is Gemini Omni 1.2 Flash at 1540, and Google is moving video generation in the Gemini app toward Gemini Omni. Veo 3.1 remains the pro route — in Google Flow, Google Vids, and Google Cloud — but even Google’s own lineup is moving on.
You can reach Veo 3.1 three ways. The consumer route is Google Flow, Google’s AI filmmaking tool, and the Gemini app, where Google AI Pro includes a limited Veo 3.1 Lite trial and Ultra includes full access. Google Vids, the Workspace video app, uses it for work videos. The developer route is Google Cloud’s Agent Platform (formerly Vertex AI), which bills per second of generated video. One route closed recently: Google removed the Veo 3.1 preview models from the Gemini API in October 2026, so developers should not plan around the Gemini API for Veo anymore.
Sora vs Veo: The Full Comparison
Here is the comparison most ranking pages get wrong, because they compare the two models as if both were still for sale. Read the first row first — it decides everything else.
| Feature | Sora 2 | Veo 3.1 |
|---|---|---|
| Available today | No — discontinued April 26, 2026 (app) and September 24, 2026 (API) | Yes — Flow, Gemini app, Vids, Google Cloud |
| Developer API | Videos API shut down; no replacement named | Google Cloud Agent Platform (formerly Vertex AI); Gemini API preview models removed October 2026 |
| Max resolution | 1080p | 4K on 8-second clips; 1080p and 720p on longer work |
| Clip length | 10 seconds by default; up to 25 seconds with Sora 2 Pro and Storyboard | 4, 6, or 8 seconds per generation |
| Native audio | Yes — speech, sound effects, ambient sound | Yes — dialogue with lip-sync, sound effects, ambience, 48kHz stereo |
| Aspect ratios | 16:9, 9:16, 1:1 | 16:9, 9:16, 1:1, 21:9 |
| Reference images | Characters — insert a person, animal, or object from a recording | Ingredients to Video — up to three reference images per clip |
| Text in video | Strong — one independent test measured 84% text-rendering readability | Weaker — the same test measured 41% readability |
| LMArena score (Nov 9, 2026) | Sora 2 Pro 1367 (#13); Sora 2 1342 (#19) | Veo 3.1 (Audio) 1363 (#14); Veo 3.1 Fast (Audio) 1361 (#16) |
| Watermarking | C2PA content credentials | SynthID in every frame and in the audio, C2PA credentials signed by Google, optional visible watermark |
| Commercial use | Moot — the product is gone | Allowed on paid plans, subject to Google’s terms and content policies |
How Much Does Veo 3.1 Cost?
Sora has no price anymore, so the only pricing that matters is Veo’s — and it has two very different shapes. Consumers pay a monthly subscription for Flow credits; developers pay per second of generated video. Both changed in 2026, and most roundups still quote the old numbers.
| Route | Price | What you get |
|---|---|---|
| Google AI Plus | $4.99/month | 200 Flow credits a month — enough to try Veo, not to work with it |
| Google AI Pro | $19.99/month | 1,000 Flow credits a month — about 50 Fast clips or 10 Quality clips |
| Google AI Ultra 5x | $99.99/month | 10,000 Flow credits a month; Fast and Quality clips cost half the credits |
| Google AI Ultra 20x | $199.99/month | 25,000 Flow credits a month; highest limits across Google AI |
| Google Cloud API — Standard | $0.40/second with audio (720p/1080p); $0.60/second at 4K | One 8-second clip costs $3.20, or $4.80 at 4K |
| Google Cloud API — Fast | From $0.10/second | One 8-second clip costs $0.80 |
| Google Cloud API — Lite | From $0.05/second | One 8-second clip costs $0.40; no 4K output |
A worked example makes the subscription math concrete. On Google AI Pro at $19.99 a month, a Veo 3.1 Fast generation costs 20 Flow credits and a Quality generation costs 100, so your 1,000 monthly credits buy about 50 Fast clips or 10 Quality clips — roughly $0.40 per Fast clip or $2.00 per Quality clip. On the API, the same 8-second Quality clip costs $3.20 at 1080p with audio, so the subscription is the cheaper route unless you need programmatic access or 4K. Two traps to avoid: first, the old Google AI Ultra price of $249.99 a month is dead — Google replaced it at I/O 2026 (May 19, 2026) with the two tiers above, $99.99 and $199.99, and roundups still quoting $249.99 are pricing a plan that no longer exists. Second, Flow credits do not roll over month to month, and video generation can be unavailable at peak hours on the lower plans, so do not plan a deadline around a same-day generation.
What Should Former Sora Users Do?
It depends on what you used Sora for. If you made clips in the Sora app, step one is the export: go to sora.chatgpt.com/sunset, click Export, and download your videos when the email arrives, because OpenAI will permanently delete Sora data after the export window closes. Step two is picking a replacement by job:
- Quick social clips and experiments: Veo 3.1 in the Gemini app or Flow on Google AI Pro ($19.99/month). The Fast tier at 20 credits a clip is the closest thing to Sora’s old casual pace, and Ingredients to Video replaces most of what Characters did, minus the social feed.
- Polished ads and client work: Veo 3.1 Quality in Flow, or the Google Cloud API at $0.40 a second with audio. The 4K option and native dialogue are the reasons to pay for Quality over Fast, and the SynthID watermark keeps the output verifiable.
- You live in ChatGPT and want video there: ChatGPT has no video model of its own anymore — no ChatGPT plan, Free, Go, Plus, or Pro, lists video generation. What works is a plugin or an MCP server: Runway or Pika for generated clips, HeyGen or Synthesia for presenter video, or a Veo or Kling connection through MCP. ChatGPT plans the video and directs the tool; the tool renders it.
- You build video into a product: the Google Cloud Agent Platform is the supported API route for Veo 3.1. Do not build on the Gemini API for Veo — Google removed the Veo 3.1 preview models there in October 2026 — and do not build on Sora at all: the Videos API is gone with no replacement named.
One more thing worth knowing before you commit to Veo: it is not the newest Google video model anymore. Gemini Omni 1.2 Flash leads LMArena’s text-to-video board at 1540, well ahead of Veo 3.1’s 1363, and Google is moving video generation in the Gemini app toward Gemini Omni. For casual generation inside the Gemini app, that is where Google is heading. For directed, pro-quality work in Flow and on Google Cloud, Veo 3.1 is still the model Google sells — just do not be surprised if the Gemini app keeps steering you toward Omni. If you want a broader view of AI video tools before you pay for anything, my guide to AI video generators and the walkthrough on how to use AI to create a video cover the wider field.
Watermarks and Provenance: Can Anyone Tell Your Video Is AI?
Yes — and that is by design on both sides. Every Veo video carries three marks: an invisible SynthID watermark embedded in the pixels of every frame and in the audio, C2PA Content Credentials digitally signed by Google LLC in the file, and an optional visible watermark you can toggle in the Gemini app. SynthID survives cropping, filters, frame-rate changes, and lossy compression, which is the point: platforms like YouTube can detect the video as AI-generated long after it leaves Google’s tools. Google opened its SynthID Detector to everyone in October 2026, so anyone can upload a clip and check it — my guide to the SynthID Detector shows how. Sora videos carried C2PA content credentials too, so the provenance habit is not new; Veo just adds the pixel-level watermark on top. If you publish AI video for clients, keep the original MP4 files alongside your edited masters and do not strip the marks — removing a watermark to pass synthetic footage off as real is the fastest way to lose a client and, under tightening rules, to face legal trouble.
The Verdict
Sora vs Veo is no longer a comparison — it is a changing of the guard. Sora 2 was a genuinely good model, effectively tied with Veo 3.1 on blind human votes, with better text rendering and a Characters feature nothing else matched. It died anyway, because it cost about $1 million a day to run for an audience that had shrunk below half a million users. Veo 3.1 wins by default and on merit: it is live, it has native audio and 4K, it costs $19.99 a month on Google AI Pro or $0.40 a second on the API, and its watermarking is the most thorough in the business. If you are choosing a video generator today, choose Veo 3.1 — and keep an eye on Gemini Omni, because Google’s next video model is already topping the leaderboard.
How this comparison was put together: Sora is discontinued and Veo 3.1 requires a paid Google AI plan, so no live generation test was possible for either model in this run, and I am not presenting any clip scores as my own testing. The verdict is built from OpenAI’s own pages (the Sora discontinuation notice, the Sora 2 launch post, and the API deprecations page), Google DeepMind’s Veo 3.1 page, Google’s pricing pages and the Google Cloud pricing page, and the LMArena text-to-video leaderboard published November 9, 2026 — all checked on November 19, 2026. Prices and plan limits change often; check the vendor’s page before you pay.
Frequently Asked Questions
No. OpenAI discontinued the Sora web and app experiences on April 26, 2026, and removed the Sora 2 models and the Videos API on September 24, 2026. OpenAI’s own help page states both dates, and the Sora 2 launch post now carries a notice that the product is no longer available. No ChatGPT plan, credit pack, or API call buys Sora video today.
Go to sora.chatgpt.com/sunset and click Export. OpenAI emails you when your content is ready to download. Do it soon: OpenAI says it will permanently delete data associated with your use of Sora after the discontinuation and after any final export window passes.
On quality, they were essentially tied. On LMArena’s text-to-video leaderboard published November 9, 2026, Sora 2 Pro scored 1367 and Veo 3.1 with audio scored 1363 — blind voters picked each about as often as the other. Sora 2 was stronger at rendering text inside video; Veo 3.1 reaches 4K and has the deeper audio. On availability, Veo wins outright, because it is the only one you can actually use.
For consumers, Google AI Pro costs $19.99 a month and includes 1,000 Flow credits — about 50 Fast clips or 10 Quality clips. Google AI Ultra starts at $99.99 a month with 10,000 credits. For developers, the Google Cloud API charges $0.40 per second with audio at 720p or 1080p, $0.60 per second at 4K, $0.10 per second for Fast, and $0.05 per second for Lite. The old Ultra price of $249.99 a month no longer exists.
Not on its own. Since Sora closed, no ChatGPT plan lists video generation. ChatGPT can still make videos through plugins — Runway or Pika for generated clips, HeyGen or Synthesia for presenter video — or through an MCP server connected to Veo, Kling, or HeyGen. ChatGPT plans the video and directs the tool; the connected tool renders it.
Veo 3.1 is the closest direct replacement for Sora’s quality, and the one this guide recommends. If you need a different job done, Kling and Runway are strong for cinematic clips, Seedance is the current value pick on the API, and HeyGen or Synthesia cover presenter-style video. But for a general-purpose text-to-video replacement with native audio, Veo 3.1 in Google Flow is the straightforward answer.