Sora API alternatives for video generation after the shutdown

It depends on which model you want to move to and how you want to buy it. OpenAI shut the Sora API down on September 24, 2026 and lists no replacement, so the alternative is another video API, chosen by five things: the model family, whether you need native audio, clip length and resolution, the API shape (one vendor's own API or a multi-model host), and the price unit and limits. Runway's API, Kling's API, fal, Replicate and Sume all run video as an asynchronous job, but they serve different models and bill in different units.

If you are weighing Google's Veo 3.1, Kling 3.0, Runway Gen-4.5 or Seedance 2.0, the table below shows where each one can be called. Veo is not in Sume's catalog, and neither is Sora. The shutdown dates come from OpenAI's help center and Deprecations page, each vendor's facts from its own pages, and the Sume side from the Video Generation docs and catalog code, all read on 2026-09-29. For hosts only, see alternatives to fal.ai and Replicate for video.

What did OpenAI shut down, and when?

OpenAI's help center dates the end of the Sora web and app experiences to April 26, 2026, and the Sora API to September 24, 2026. OpenAI notified developers on March 24, 2026. Its Videos API reference now says the models and the API are no longer available, that no one-to-one replacement API is available, and that the page is kept for historical reference.

Sora 2 API pricing lists the retired model ids and what each clip cost, and Seedance vs Sora 2 sets the old specs beside the Seedance ids.

What should you check before picking a Sora API alternative?

Start from what your Sora integration relied on. Its create call took 4, 8 or 12 seconds and four sizes up to 1792x1024, so check each option on these points:

• **Models:** does the API serve the model family you want, or only its own?
• **Audio:** does the model generate a sound track with the video, and does that change the price?
• **Length and resolution:** the clip lengths and resolutions each model accepts.
• **API shape:** async submit and poll, a webhook, idempotent retries, and one request body across models or one per model.
• **Price unit and limits:** per second, per video, per credit or per GPU second, plus concurrency caps and how long results stay downloadable.

What are the alternatives to the Sora API?

Five video APIs, in alphabetical order. Each row uses only that company's own pages; prices are examples of the unit, not a full price list.

Where can you call Veo, Kling, Runway and Seedance models?

Most Sora replacements are a model family first and a vendor second. This table shows which of the five APIs list each family on the pages read. A dash means the page read did not list it.

Which alternative fits which job?

Start from the model and audio you need, then from how you want to be billed. Web apps and editors are a separate choice; every option here is an API.

• **Runway's API** if you want gen4.5 or Veo 3.1 behind one key. Request bodies differ per model there, so the valid ratio and duration depend on the model you pick. Its Model Router saves a routing configuration so you don't hardcode a model id.
• **Kling's API** if Kling models are all you need, including what its capability map lists per model: multi-shot generation, native audio, first and last frames, and 4K on Kling 3.0.
• **fal or Replicate** if you want one account for many hosted models, including Veo 3.1, or to run your own weights: fal prices GPUs for custom deployments, and Replicate documents deploying custom models.
• **Sume** if you want one request body, idempotent submits, and one USD balance across the Kling, Seedance and other families in its catalog. One API for Kling, Seedance and other video models covers that setup.
• **Not Sume** if you need Veo, Runway's gen4.5, a Sora model, clips over 30 seconds, or your own weights.

Where does Sume fit, and does it have a Sora model?

Sume has no Sora or Veo model. The ids its video catalog lists are seedance-2.5, seedance-2-mini, seedance-2, seedance-2-fast, kling-3, wan-3.0, grok-imagine-video-1.5, minimax-h3, minimax-h3-max, gemini-omni-flash-1.1, h3-max-recast. You can also send sume/auto and let Sume pick: the response echoes sume/auto, and Sume never discloses which family ran.

Limits are per model. seedance-2.5 and wan-3.0 accept up to 30 seconds and gemini-omni-flash-1.1 3 to 10 seconds, and the others top out at 15 seconds. gemini-omni-flash-1.1 lists 4K, and seedance-2.5 and seedance-2 offer 1080p. generate_audio asks for a sound track on models that can make one; minimax-h3-max and gemini-omni-flash-1.1 document native audio.

Prices come from each model's pricing_skus on GET /v1/videos/models. As one example, kling-3 is billed per second: $0.14 silent, $0.21 with sound, plus a 5.5% agent fee by default. How much an AI video costs has every model's rate, and listing video models walks through the capability fields.

How do the old Videos API fields map to POST /v1/videos?

Sume does not take the old request body as is. Its OpenAPI schema for POST /v1/videos has no seconds or input_reference field and refuses fields it does not list, so rename each field:

What does the migrated request look like?

This is the docs' sume/auto request with an idempotency key added. Swap in a catalog id to pin a model.

curl -X POST "https://api.sume.com/v1/videos" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: clip-001" \
  -d '{
    "model": "sume/auto",
    "prompt": "A vertical UGC-style product clip on a desk, natural light",
    "aspect_ratio": "9:16",
    "duration": 5
  }'

What else changes when I switch?

The request shape is not the only difference. Plan for these on day one:

• Auth is Authorization: Bearer $SUME_API_KEY, sent from your server.
• Send an Idempotency-Key: a replay with the same key and the same body returns the original job instead of a second one.
• For push instead of polling, pass an HTTPS callback_url. Sume posts its own job webhook envelope, signed with x-sume-webhook-signature.
• The same job is also visible at GET /v1/jobs/{id}/status and GET /v1/jobs/{id}/result.
• Downloads from GET /v1/videos/{id}/content need your API key and answer with a redirect to the file, so use curl -L.
• On 429, back off and use retry-after when present. queue_full means the workspace can't take another paid job until a queued or running one finishes or is canceled.
• Billing draws on your workspace USD balance, reserved on submit, and usage.cost on the poll response is the billable amount.