Kling 4.0 API: what to generate with on Sume while you wait

If you searched for a Kling 4.0 API, the answer today is kling-3. Kling 4.0 was announced in September 2026, and Sume's video catalog lists kling-3; this post gives no Kling 4.0 specs or prices because there is nothing to verify yet. What you can do now is run real jobs on kling-3 and keep the same request shape for whatever the catalog lists next.

On Sume, kling-3 takes 4 to 15 seconds at 720p or 1080p, in 16:9, 9:16 or 1:1, and bills $0.14 per second with audio off and $0.21 per second with audio on, both already including the provider list price times 1.25.

What a kling-3 clip costs today

Audio is the only price switch. generate_audio defaults to on, so a request that omits it is billed at the audio rate.

The request

It is the standard asynchronous /v1/videos call from Video generation. Poll polling_url until the status is completed, then download unsigned_urls[0].

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: kling3-wait-001" \
  -d '{
    "model": "kling-3",
    "prompt": "A runner laces shoes at dawn, handheld, shallow focus",
    "duration": 8,
    "generate_audio": false
  }'

How to stay ready

Do not wire a model id into your code by hand. Read GET /v1/videos/models at startup and pick from the list, so a new catalog entry is an id swap rather than a rewrite.

• Keep prompts in files, one per scene, so they can be re-run on any id later.
• Send an Idempotency-Key so a retry returns the original job.
• Budget with the numbers above: the hold at submit is the full price of the request.

What to do this week

Treat the announcement as a reason to prepare, not a reason to pause. Prompts, reference images, naming and cost tracking are all model-independent work, and a 5-second kling-3 clip with sound is $1.05, so a full day of experiments is cheap.

Sume reserves the full price at submit, so the number in the table is also the amount that must be free in the balance when you send the request. If it is not, the call fails with 402 insufficient_credits before any provider work starts.

If a shot needs more than 15 seconds, kling-3 cannot do it in one job. wan-3.0 and seedance-2.5 accept up to 30 seconds, and Timeline can join separate takes.