Sora 2 Review 2026: What It Did Best, and Why It’s Gone

Correction: this review was published on May 1, 2026, five days after the app closed, and originally said Sora 2 had moved into ChatGPT Pro. That was wrong — OpenAI's discontinuation notice describes no successor product — and the access and pricing sections were rewritten on September 8, 2026. The hands-on assessment of the model, made while it was available, is unchanged.

The Bottom Line

Sora 2 was the most physically accurate and stylistically capable AI video model of its era. After three months of intensive use across marketing, social media, and concept work, it changed our content production pipeline — which is exactly why its disappearance matters. It is no longer available to consumers; developers can call the API until September 24, 2026, and after that the model is gone. What follows is our record of what it did well, where it fell short, and what to use instead.

What Sora 2 Did Best

Physics and Materials

Water, smoke, fabric, hair — Sora 2's understanding of how the physical world moves is genuinely impressive. We tested edge cases (gymnastics routines, glass shattering, water splashing in slow motion, leaves blowing through doorways). Sora 2 produces believable results 75-85% of the time. No other video model approaches this physics fidelity.

Synchronized Audio

The headline feature. Sora 2 generates dialogue, sound effects, and ambient audio in sync with the video. No more silent AI clips. A fully realized audiovisual scene from a single prompt.

Camera Movement

Specify "slow dolly in", "tracking shot from the right", "drone pull-back over forest" and Sora actually delivers cinematographic camera language. The camera moves feel intentional and stable, not jittery or uncanny. This is where it leaves competitors clearly behind.

Storyboards Mode (Pro tier of the Sora app)

The most underrated feature. Drop "prompt cards" along a 25-second timeline, specifying what should happen at second 2, second 5, second 8. Sora 2 generates with progression in mind. Dramatically improves narrative coherence over single-prompt generation.

Identity Insertion

By observing a video of someone, Sora 2 can insert them into any Sora-generated environment with accurate appearance and voice. Works for humans, animals, or objects.

In-App Editor

Trim clips with frame-level precision, stitch multiple clips into a sequence, reorder. Available on iOS and web.

Specs and Limits (as shipped)

• All app users: 1080p, 15-second clips
• Pro tier: 25-second clips with Storyboards, no watermark
• Generation time: ~1-3 minutes per clip
• API, until September 24, 2026: sora-2 and sora-2-pro, 16- and 20-second generations, 1280x720 / 1920x1080 / 1080x1920 / 480p output, extendable in 20-second steps to 120 seconds (OpenAI API docs)

Where It Still Struggles

Long-Form Continuity

Across multiple clips meant to feature the same character, identity drift remains significant. For narrative content with recurring characters, expect substantial editing or planning around the limitation.

Hands and Fine Motor Actions

Better than image AI but still imperfect. Complex hand gestures, instrument playing, precise object manipulation occasionally produce uncanny results.

Complex Cause and Effect

Multi-step physical sequences sometimes produce impossible results. Single-action shots are reliable; "person bites apple, apple has bite mark" sometimes fails.

25-Second Cap

No native long-form generation. Longer projects require stitching multiple clips, with associated continuity challenges.

Availability and Pricing (September 2026)

There is nothing left to buy as a consumer. The Sora app and web experience closed on April 26, 2026, and Sora 2 is not part of ChatGPT Plus or Pro. Until September 24, 2026 the only route is the API — sora-2 for speed, sora-2-pro for quality, at OpenAI's published API rates — after which the models and the Videos API are removed with no replacement. Purchased ChatGPT/Sora credits can still be spent on Codex, and existing videos can be exported at sora.chatgpt.com/sunset before OpenAI deletes account data.

What to Use Instead: Veo, Kling, Runway

Sora 2 wins: physics accuracy, synchronized audio, cinematic quality.

Google Veo 3 wins: Workspace integration, longer clips on Ultra plan, 4K output.

Kling wins: price, speed, character consistency for talking heads.

Runway wins: in-browser editing studio, Motion Brush, mature post-production tools.

For migrating projects: Google Veo 3.1 is the closest like-for-like on cinematic quality and audio, Kling is the volume-and-price pick, and Runway Gen-4.5 is the production-and-editing standard. xAI's Grok Imagine 1.5 has since entered image-to-video as well.

Tips & Best Practices

• Generate 3-4 variations of important shots — quality varies
• For ads/marketing, lead with the camera move ("slow dolly in to a steaming coffee cup...")
• Use Storyboards mode for narrative content — much better consistency than single prompts
• Reference cinematographers or films ("Roger Deakins lighting", "shot like Blade Runner 2049")
• Edit Sora outputs in DaVinci Resolve or Premiere — color grading + sound design transforms results
• For longer projects, plan as 25-second beats, then stitch in your editor

Verdict

Sora 2 redefined what was possible in AI video, and then OpenAI switched it off. For storytelling, marketing teasers, concept work, and any scene where realistic physics mattered, it was the best option available while it lasted — which is what our 4.6 recorded, and why the score stands as a historical verdict rather than being revised for a product you can no longer buy. If you are choosing a video model today, start with Veo 3.1. Score (historical, May 2026): 4.6/5.

Frequently Asked Questions

Is Sora 2 still available?

No, not to consumers. OpenAI discontinued the Sora app and website on April 26, 2026. The API remains until September 24, 2026, after which the Sora 2 models are removed with no replacement.

Can I use Sora 2 through ChatGPT Plus or Pro?

No. OpenAI's discontinuation notice describes no successor product and no ChatGPT integration. Purchased ChatGPT/Sora credits can be used for Codex instead.

How much does Sora 2 cost now?

Nothing is for sale to consumers. Developers can still call sora-2 and sora-2-pro through the API at OpenAI's published rates until September 24, 2026.

How long could a Sora 2 video be?

On the Sora app, 15 seconds for all users and 25 seconds on the Pro tier using Storyboards. The API supports 16- and 20-second generations, extendable in 20-second steps to a maximum of 120 seconds.

How do I export my Sora videos before they are deleted?

Go to sora.chatgpt.com/sunset and click Export; OpenAI emails you when the export is ready. OpenAI says it will permanently delete Sora data after any final export window closes.

What should I use instead of Sora 2?

Google Veo 3.1 is the closest like-for-like on cinematic quality and synchronized audio, Kling is the volume-and-price pick, and Runway Gen-4.5 is the production-and-editing standard. Our guides cover all three.

Is the Sora API being discontinued?

Yes. OpenAI planned to discontinue the Sora API on September 24, 2026. After that date there is no way to generate Sora 2 video at all — OpenAI's deprecation table lists no replacement.

Does Sora 2 generate audio?

Yes. Sora 2 generates dialogue, sound effects and ambient audio in sync with the video from a single prompt, producing a fully realised audiovisual scene rather than a silent clip.

What is Storyboards mode in Sora 2?

A Pro-tier feature that lets you drop prompt cards along a 25-second timeline, specifying what should happen at particular seconds. Sora 2 then generates with that progression in mind, which dramatically improves narrative coherence compared with a single prompt.

Is Sora 2 better than Veo 3, Kling or Runway?

It depends on the job. Sora 2 wins on physics accuracy, synchronised audio and cinematic quality. Google Veo 3 wins on Workspace integration, longer clips on the Ultra plan and 4K output. Kling wins on price, speed and character consistency for talking heads. Runway wins on in-browser editing tools. Most production teams use more than one.