Sora Is Shutting Down: What Happened and Where to Go Next

OpenAI has retired Sora. The consumer app is already gone, and the API follows this month. If you still have Sora in a product or a workflow, the migration window is closing.

The Two Shutdown Dates

There are two separate events, and only one has already happened.

The API retirement covers the sora-2 and sora-2-pro models and the whole /v1/videos surface: generation, retrieval, remix, extend, download, and delete. OpenAI's developer documentation now marks the video API as deprecated and lists no replacement video API.

Sources: What to know about the Sora discontinuation and the OpenAI video API reference.

What Sora Was

Sora was OpenAI's text-to-video model, first shown in February 2024 and opened more widely with Sora 2 in 2025. It generated video from a written prompt, and later from a still image, with synchronized audio in its second generation.

The pieces that made it notable:

• Prompt-to-video in one step. No storyboard, no keyframes. A paragraph of description produced a moving shot.
• Physical plausibility. Sora 2 was the first widely available model where objects, water, and crowds mostly behaved as expected across a shot.
• Native audio. Dialogue, effects, and ambience came out of the same generation rather than a separate pass.
• Remix and extend. An existing clip could be re-prompted or lengthened through the API.

It also set the public expectation that a single prompt should produce a finished, sounded clip. Every model in this space is now measured against that.

What the Shutdown Means for You

If you used the Sora app: it is already gone. Export windows for app-created content have largely closed, and OpenAI has said Sora data is eventually deleted permanently.

If you use the Sora API in production: you have until September 24, 2026. After that date the endpoints stop responding. There is no drop-in OpenAI replacement, so this is a provider migration, not a version bump.

Practical checklist:

1. Export any assets you still need. Download finished renders now rather than relying on retrieval endpoints that are about to disappear.
2. Inventory your call sites. Anything hitting /v1/videos needs to move, including remix and extend flows.
3. Re-test prompts on the new model. Prompt behaviour is not portable between video models; the same text produces different framing and pacing.
4. Expect schema changes. Duration, aspect ratio, and audio fields differ per provider, and the values are not interchangeable.

Where to Go Next

The gap Sora left has been filled quickly, and by more than one model. Which one fits depends on what you were using Sora for.

Replacing prompt-to-video with sound: Veo 3.1 is the closest match on visual fidelity and prompt adherence, with audio-aware output. Grok Imagine Video 1.5 also generates native sound and holds fine physical detail.

Replacing long single takes: Sora's shot length was a real constraint. Seedance 2.5 and Wan 3 both reach 30 seconds, with Wan 3 outputting 1080p with sound.

Replacing image-to-video: Kling O3 animates a still and can control the end frame as well as the start, which Sora could not do directly. Wan 2.7 offers the same first/last-frame control at 1080p.

Replacing iterative editing: Gemini Omni supports multi-turn language-driven edits with scene consistency — closer to Sora's remix flow than a plain re-generate.

If you need maximum resolution: MiniMax H3 reaches 2K and accepts long prompts with mixed references.

The Honest Caveat

No single model is a Sora clone. The differences that matter in migration:

• Duration ceilings vary, from a few seconds to 30.
• Audio is not universal. Some models generate it natively, some not at all, and the request field is named differently on each.
• Aspect ratio support differs. Not every model accepts 1:1 or 21:9.
• First/last-frame control is model-specific, and the parameter names are not shared.

That is why comparing on your own prompt matters more than reading a spec sheet. The same brief renders differently on each model.

Try the Alternatives Side by Side

SoraRemix runs the ten models above through one interface, so you can put your actual Sora prompt into several of them and compare the output before committing to a migration.

1. Browse the model directory and pick the two or three closest to your old use case.
2. Paste the prompt you were running on Sora.
3. Compare duration, audio, and framing on the same brief.

If you want a starting point, our roundup of free options covers which models you can test without paying upfront.