One of these two models is no longer on sale, and OpenAI says so on its own page. The Sora 2 announcement now opens with a notice: “As of April 26, 2026, the Sora product is no longer available.”
Google’s Veo 3.1 page carries no such notice. It is still the model Google DeepMind introduces as “Video, meet audio”.
The short answer: one model is sold, the other is retiring
Veo 3.1 is Google’s video generation model with native audio, documented at eight seconds per clip in 1080p and 4K. Sora 2 was OpenAI’s video-and-audio model, closed since April 26, 2026.
Anyone picking a tool for work that starts next month has finished deciding. The rest of this page puts the two vendors’ own pages side by side and reports what they print.
What each vendor’s own pages confirm, side by side
| Published item | Veo 3.1 · Google | Sora 2 · OpenAI |
|---|---|---|
| Model page | Google DeepMind’s Veo page | OpenAI’s Sora 2 announcement |
| Shutdown notice | None | “no longer available” since April 2026 |
| Clip length | “Veo videos are 8 seconds long.” | No duration stated |
| Resolution | 1080p and 4K | Not stated |
| Native audio | Generated with the picture | “synchronized dialogue and sound effects” |
| Character consistency | Reference images, first-and-last frame | Not documented there |
| Extending a clip | Documented | Not published |
| Price | None | None |
Four of those rows exist only because one vendor chose to publish them. Read the table as a list of what is verifiable, not as a scoreboard.
Availability comes before any spec row
OpenAI’s pages give dates, not opinions. The app and web experiences were discontinued on April 26, 2026, and the API follows on September 24, 2026 — the date that matters if you maintain an integration.
Two practical answers sit on the same page. Export still works: “We recommend you export your Sora content as soon as possible.” Bought credits are not lost, since “any purchased ChatGPT/Sora credits can still be used for Codex”.
Google has published nothing comparable for Veo 3.1, because there is nothing comparable to publish.
Published specifications: one vendor prints numbers
Google’s model page is specific. “Veo videos are 8 seconds long.” Resolution is spelled out there too: 1080p for editing, 4K for high-end production. The documented capabilities are concrete names you can look for in the interface: character consistency from reference images, scene extension, camera controls, motion controls, first-and-last-frame transitions, outpainting, ingredients to video.
Google prints a limitation on the same page, worth more than any list of strengths: natural and consistent spoken audio, especially for shorter speech, remains an area of active development.
Where the fifteen-second figure comes from
Search this comparison and you will meet durations for Sora 2 — fifteen seconds for regular users, twenty-five for a Pro tier — attributed to an announcement. OpenAI’s page prints neither. The only durations on it sit inside a sample creative brief, where “5.0s” and “4K” describe one shot the poster wrote.
Sound: both vendors claim native audio
Audio is the one capability both pages describe in similar language. Google says Veo 3.1 generates sound effects, ambient noise and dialogue natively with the picture. OpenAI calls Sora 2 a general purpose video-audio system able to build soundscapes, speech and effects with a high degree of realism.
Google then publishes benchmark charts where its own model wins on overall preference, prompt adherence and audio-video alignment. Those are the vendor’s own results from the vendor’s own benchmarks, not independent testing.
Price: neither model page prints a number
Neither page contains a price — no subscription figure, no credit balance, no per-second rate. Google keeps its published numbers elsewhere, and we work through them in the price and free access guide. The developer-side rates sit in the API and Vertex AI pricing guide.
OpenAI’s announcement said Sora 2 would “initially be available for free, with generous limits to start”, and that the only plan discussed was charging for extra videos when demand outran compute. No amount was attached to that sentence.
Worked example: one prompt, two pipelines
Take a prompt that needs a spoken line and a camera move inside one short clip.
The prompt
“A medium shot of a baker in a flour-dusted apron pulling a tray of bread from a stone oven. Steam rises into warm window light from the left. The camera drifts slowly right. She looks up and says: ‘Careful, it’s hot.’ Audio: an oven door creaking, bread crust crackling, a low kitchen hum.”
Parameters: eight seconds, 1080p, 16:9, audio on, one generation.
What comes back on each side
On Veo 3.1 the prompt maps onto documented behaviour. Eight seconds fits one shot with one line, and the dialogue and audio bed are generated with the picture. If the action will not fit, the documented next step is to extend the scene from the last second of the clip, keeping audio consistent.
On the OpenAI side the same prompt has nowhere to go: the app closed on April 26, 2026, and API access ends on September 24, 2026.
We have not run the two side by side. The prompt guide lists the elements Google documents; the getting-started walkthrough covers a first clip end to end.
What neither vendor has published
OpenAI never published a duration specification, a resolution list, an extension capability or any rate for Sora 2 on the pages checked here, and its developer pricing documentation was not reachable during this check.
Google’s Veo 3.1 model page publishes no price, no subscription tier, no credit balance, no free allowance and no conversion from Flow Credits into generations. Watermark policy is absent too. Without a published conversion rate, nobody can turn a credit balance into a clip count — and any page that does the arithmetic for you has supplied the missing number itself.
Frequently asked questions
Can I still use Sora 2? The app closed on April 26, 2026; the API stops on September 24, 2026.
Is Veo 3.1 available? Yes. Google’s model page is live and carries no shutdown notice.
Which one makes longer clips? Google publishes eight seconds and documents extension. OpenAI publishes no duration at all.
Which is cheaper? Neither model page prints a price, so this cannot be answered from them.
Where should a beginner start? On the model still on sale — the topic hub collects the tutorial, the prompt guide and the developer documentation.
A comparison built on first-party pages is shorter than one built on borrowed numbers, and a good deal more useful. Check availability first, then read what the vendor actually printed.