Veo 3.1 vs Sora 2: what changed after Sora shut down

Veo 3.1 and Sora 2 are no longer a choice: OpenAI removed the Sora 2 API on 24 September 2026 and named no replacement. Veo 3.1 is still sold: $0.40 per second with audio on Runway's API, so an 8-second clip is $3.20 (2026-10-26). If you need a Sora-style model today, use Veo 3.1 or another live model.

Key facts

Sora 2 API removal24 September 2026 (Videos API, sora-2, sora-2-pro and dated snapshots); no replacement listed[1]
Sora 2 clip length and size (API guide)4, 8, 12, 16 or 20 seconds; 720p on sora-2, up to 1080p on sora-2-pro[2]
Veo 3.1 clip length and size (Google, 16 Oct 2025)4, 6 or 8 seconds; 720p or 1080p; 16:9 or 9:16[4]
Veo 3.1 on Runway's API40 credits/s with audio, 20 without; $0.01 per credit[6]
Veo 3.1 Fast on Runway's API15 credits/s with audio, 10 without[6]
Veo 3.1 preview models in the Gemini APIShut down 22 October 2026; Google points to Gemini Omni Flash or the GA models[3]

Most Veo 3.1 vs Sora 2 articles were written while both models could be bought. One of them can't now. This page states what is still live, sets the documented specs side by side, shows Google's head-to-head benchmark and prices a Veo 3.1 clip.

Is Sora 2 still available?

Sora 2 is retired in OpenAI's API: the Videos API, sora-2, sora-2-pro and their dated snapshots were removed on 24 September 2026.

OpenAI's deprecation table says developers were notified on 24 March 2026 and lists no replacement model. The Decoder reported that the Sora web and app closed on 26 April 2026, and that OpenAI wants to put its compute into coding tools and enterprise customers. Any comparison that still lets you pick Sora 2 is out of date.

Veo 3.1 vs Sora 2: documented specs side by side

Sora 2 allowed longer clips, and only Veo 3.1 can still be bought. Status as of 2026-10-26.

Sora 2 Veo 3.1
Status Retired. API removed 24 Sep 2026 Live. Gemini API preview IDs shut down 22 Oct 2026; Google points to GA models
Clip length 4, 8, 12, 16 or 20 seconds 4, 6 or 8 seconds
Resolution 720p (sora-2); up to 1080p (sora-2-pro) 720p or 1080p
Aspect ratio Portrait and landscape sizes 16:9 or 9:16
Audio Dialogue and sound effects with the video Dialogue, SFX: and Ambient noise: in the prompt
Reference input One image, plus up to two character IDs Reference images for scene, character, object or style
Extension Up to 6 extensions, 120 s total Extend continues a clip; Google says a minute or more
Likeness Cameos with owner permission No cameo system documented

Sources: OpenAI's Sora 2 prompting guide and deprecations table, Google's Veo 3.1 blog posts from 15 and 16 October 2025, and Google's Gemini API video page. Google's clip-length and resolution figures date from October 2025.

Veo 3.1 vs Sora 2: which was better?

Google's benchmark says Veo 3.1. It is maker-run, so treat it as a signal and test your own prompt.

Google's Veo page (last updated October 2025) compares Veo 3.1 with Sora 2 Pro on 1,003 MovieGenBench text-to-video prompts and reports Veo 3.1 best on overall preference. Launch coverage said Sora 2 added synchronized dialogue and more physically accurate motion; Google said Veo 3.1 brought richer audio, more narrative control and reference images. Sora 2 is retired, so the practical test now is Veo 3.1 against any other live model on your prompt.

What does Veo 3.1 cost per clip?

On Runway's API, one Veo 3.1 clip of 8 seconds costs $3.20 with audio and $1.60 without, at list price on 2026-10-26.

Runway sells credits at $0.01. Veo 3.1 is 40 credits per second with audio and 20 without. Veo 3.1 Fast is 15 and 10.

Tier Rate 8-second clip 3 tries per keeper
Veo 3.1, audio $0.40/s 8 × $0.40 = $3.20 $9.60
Veo 3.1, silent $0.20/s 8 × $0.20 = $1.60 $4.80
Veo 3.1 Fast, audio $0.15/s 8 × $0.15 = $1.20 $3.60
Veo 3.1 Fast, silent $0.10/s 8 × $0.10 = $0.80 $2.40

The "3 tries" column is a budgeting rule of thumb. Draft on Fast, then render the chosen take on the full model: two Fast drafts plus one full render is $2.40 + $3.20 = $5.60 with audio. These figures cover Runway's API; check your chosen app's pricing separately. For wider cost sums, see the Sora alternatives price table.

How do I move from Sora 2 to Veo 3.1?

Keep the scene text, rewrite the dialogue, and split anything longer than 8 seconds into shots.

Sora 2 habit Veo 3.1 change
12 to 20 second clips Veo 3.1 stops at 8 seconds. Plan two or three shots and join them
Dialogue in its own block Put speech in quotation marks. Add SFX: and Ambient noise: lines
characters or cameo IDs Use reference images you have the right to use
size set in the API call Choose 720p or 1080p and 16:9 or 9:16 on the Veo platform
Extend up to 120 seconds Use Veo's Extend, or cut a longer story into shots

Google calls its audio and advanced controls experimental, so budget a retry or two. If a clip comes back silent, see Veo 3 no audio fixes. Our guide to Sora 2 prompts that still work has a worked rewrite.

Do Veo 3.1 and Sora 2 allow real faces and characters?

Use your own face, people who agreed in writing, or original characters.

Sora's cameos let an owner permit use of their likeness and revoke it (GSMArena). Veo 3.1 documents reference images for consistency and has no cameo system.

For other live models, see the HoneyVid learn section or the Veo and Sora guides.

Questions

Is Sora 2 better than Veo 3.1?

Sora 2 cannot be called in OpenAI's API after 24 September 2026, so Veo 3.1 is the one you can use. Google's October 2025 benchmark of 1,003 MovieGenBench prompts reports Veo 3.1 best on overall preference against Sora 2 Pro. It is a maker-run test, so run your own prompt to confirm.

Which Veo version is best, Veo 3.1 or Veo 3.1 Fast?

Veo 3.1 is the full model and Veo 3.1 Fast is the cheaper tier. On Runway's API, Fast costs 15 credits a second with audio against 40 for Veo 3.1. Draft on Fast, then render the keeper on the full model.

What is better than Veo 3?

Kling, Seedance, Grok Imagine and others are live alternatives, and our Sora alternatives guide lists their prices. Run the same prompt on two models and keep the clip that wins.

Can I still use my old Sora 2 prompts?

Yes, as text. Veo 3.1 takes plain scene descriptions, with dialogue in quotation marks and optional SFX: and Ambient noise: lines. Cut any request for more than 8 seconds into shots.

Can I use Sora cameos or real people's faces in Veo?

Use your own face or people who have agreed. Sora's cameo feature let an owner permit and revoke use of their likeness. Veo documents reference images for consistency, with no cameo system.

Sources

  1. Deprecations OpenAI. Accessed .
  2. Sora 2 prompting guide OpenAI Cookbook. Accessed .
  3. Generate videos with Veo Google AI for Developers. Accessed .
  4. Ultimate prompting guide for Veo 3.1 Google Cloud Blog, published . Accessed .
  5. Veo 3.1 and new creative capabilities in the Gemini API and Flow Google, published . Accessed .
  6. API pricing Runway. Accessed .
  7. Google DeepMind Veo model page Google DeepMind. Accessed .
  8. OpenAI releases Sora 2 video model with improved realism and sound effects GSMArena, published . Accessed .
  9. OpenAI sets two-stage Sora shutdown with app closing April 2026 and API following in September The Decoder, published . Accessed .

More on Veo and Sora

All Veo and Sora guides