Skip to contentOceanToken

5 min read

Official Kling API vs a Multi-Model Gateway

Official Kling API or a multi-model gateway? Kling versions, keys, concurrency and result links compared, plus a simple rule for which route fits your product.

By OceanToken Team

Official Kling API vs a Multi-Model Gateway

Editor's note: Kuaishou said Kling AI's annualized revenue run rate was about US$500 million in March 2026, The Standard reported . Kling is a serious product, yet most teams still run it next to Seedance, an LLM and an image model. We run a gateway, so weigh our view: Kling only? Go official. Otherwise, one key beats a separate vendor account per model.

Short answer: Use the official Kling API from Kuaishou if Kling is your only video model or you need the newest versions, such as Kling 4.0, the day they ship. Use a multi-model gateway when Kling is one of several models in your product and you want one key, one balance and one SDK.

What the official Kling API covers

Kuaishou's video capability map  lists eight video models on its own API, checked 2026-11-09:

Model Duration Resolution Multi-shot (text/image to video) Native audio
Kling 4.0 3 to 30 s 720p, 1080p, 4K Yes Yes
Kling 4.0 Flash 3 to 20 s 720p Yes Yes
Kling 3.0 Turbo 3 to 15 s 720p, 1080p Yes Yes
Kling 3.0 3 to 15 s 720p, 1080p, 4K Yes Yes
Kling 3.0 Omni 3 to 15 s 720p, 1080p, 4K No (yes under Omni Reference) No (yes under Omni Reference)
Kling O1 3 to 10 s 720p, 1080p No No
Kling 2.6 5 or 10 s 720p, 1080p Single-shot only No
Kling 2.5 Turbo 5 or 10 s 720p, 1080p Single-shot only No

That breadth is the main argument for going direct. Kuaishou ships each new Kling version on its own API, and every other route can only add it afterwards.

How access works on the official route

The quick start  is short. You create an API key in the Kling Developer Console, copy it once (it is not shown again), and send it as a Bearer token. Usage shows up in the console as call volume, resource package consumption, and unit and balance deductions.

Capacity is the part that needs planning. The concurrency rules  count running tasks per account, per model version and per package type, and the quota comes from the highest-concurrency active package of that type. Go over it and task creation fails with code 1303. Results are kept for 30 days, and you can get status by polling or through a callback_url, as described in the callback protocol . Our Kling n8n and Make workflow guide walks through that create, poll and fetch loop step by step.

None of this is hard. It is one more vendor account, one more billing relationship and one more request format, and that cost grows with every other model you add next to Kling.

What a multi-model gateway changes

A gateway puts many models behind one key and one request format. On OceanToken that format is OpenAI-compatible, with base URL https://api.oceantoken.ai/v1. On 2026-11-02, /kling-api listed four Kling ids: kwaivgi/kling-v3.0-pro, kwaivgi/kling-v3.0-std, kwaivgi/kling-v3.0-omni and kwaivgi/kling-video-o1. Kling 4.0 and 4.0 Flash were not among them.

Official Kling API OceanToken gateway
Kling versions Eight video models, newest first Four Kling ids on /kling-api
Other models on the same key Kling models only Yes: other video, image and text models
Request format Kling's own (/text-to-video/kling-3.0, settings, options) OpenAI-compatible videos API
Finding out a job is done Poll or callback_url Poll GET /v1/videos/{id}; no webhook documented
Status values submitted, processing, succeeded, failed queued, in_progress, completed, failed
Result link Kept 30 days content_url valid 24 hours
Failed job Not covered in the pages we read Not charged, hold released

The OceanToken column comes from its video guide  as of 2026-11-02. The trade is plain: the official API has more Kling and the gateway has more of everything else.

Third-party hosts in general

Beyond these two routes, several third-party hosts resell or wrap Kling with their own schemas. The same checks apply to any of them, gateway included:

  • Model names differ. A Kling id from one route will not work on another. Keep ids in config, not in code.
  • Status strings differ. Even Kuaishou has two: legacy callbacks for Kling 3.0 Omni and earlier say succeed, newer ones say succeeded. Map every route's values to one internal status in one function.
  • Versions can drift. Check that the version and resolution you asked for are the ones you got, especially after a route adds a new model.

Which route fits your setup

Diagram: Official Kling API or a multi-model gateway, by setup

  • Kling is the only model you run: the official Kling API.
  • You need Kling 4.0 or 4.0 Flash now: the official Kling API.
  • You want callbacks instead of polling: the official Kling API, which documents callback_url.
  • Kling is one of several models in your product: a multi-model gateway.

Our take: Pick by the rest of your stack, not by Kling alone. A video product that also calls an LLM and an image model pays an integration tax for every extra vendor, and in our view the cheapest-looking route can stop looking cheap once you count retries and adapter work. If Kling is the whole product, or you need a version a gateway does not list yet, go official.

How to test this on your workload

  1. Pick three prompts: a 5-second product spin, a 10-second dialogue and a multi-shot scene.
  2. Run each on both routes at the same duration and resolution.
  3. Log success rate, time to completion, output resolution and how failures are reported.
  4. Count the code you had to write per route: auth, status mapping, download and retries.

Running it on OceanToken

Set the base URL to https://api.oceantoken.ai/v1 and use kwaivgi/kling-v3.0-pro. The /kling-api example uses the OpenAI SDK's videos.create_and_poll, so an existing OpenAI client only needs a new base URL and key. Moving a job from Kling to another video model is a change of model id, and a failed video job is not charged. See the current rate on /kling-api. The Kling v3.0 Pro model page lists its resolutions and inputs.

FAQ

Is there an official Kling API? Yes. Kuaishou runs it, with keys from the Kling Developer Console and docs at kling.ai.

Which Kling versions does the official API offer? Its capability map lists Kling 4.0, 4.0 Flash, 3.0 Turbo, 3.0, 3.0 Omni, O1, 2.6 and 2.5 Turbo.

Do all routes expose the same Kling features? No. Versions, request fields, status strings and result retention differ by route.

Can I use one key for Kling and LLMs? Yes, on a multi-model gateway such as OceanToken, through one OpenAI-compatible base URL.

Sources

  1. Kling API docs: video capability map  (checked 2026-11-09)
  2. Kling API docs: quick start  (checked 2026-11-06)
  3. Kling API docs: concurrency rules  (checked 2026-11-09)
  4. Kling API docs: callback protocol  (checked 2026-11-09)
  5. OceanToken Kling API page (checked 2026-11-02)
  6. OceanToken video guide  (checked 2026-11-02)
  7. The Standard: China's Kuaishou records 26pc drop in Q1 profit, AI revenue jumps  (checked 2026-11-02)

Get your OceanToken API key

  1. Create an OceanToken account and an API key.
  2. Copy the snippet below, paste your key, and run it.
  3. Check the current rate on /kling-api.
curl https://api.oceantoken.ai/v1/videos \
  -H "Authorization: Bearer $OCEANTOKEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "kwaivgi/kling-v3.0-pro", "prompt": "A cat walks past a bakery sign at dawn"}'

Get API Key 

  • Kling
  • AI video
  • Compare
  1. 8 min read

    Seedance 2.5 vs Wan 3.0 Prime by Job

    Seedance 2.5 vs Wan 3.0 Prime for API video: what each call can hold, how each is billed, and which jobs to send to which model, with ids and code.

    • Seedance
    • Wan
    • Compare
  2. 9 min read

    Seedance vs Veo 3.1 for 1080p, by Shot

    Seedance vs Veo 3.1 for 1080p: when Seedance 2.0 Upscaled is enough, when to render native on Seedance 2.5 or Veo 3.1, and the model ids for each shot.

    • Seedance
    • Veo
    • Compare
  3. 7 min read

    Qwen API: Alibaba Direct vs a Gateway

    Qwen API from Alibaba Model Studio direct vs an OpenAI-compatible gateway: regional keys and endpoints, workspaces, default thinking and enable_thinking.

    • Qwen
    • LLM API
    • Compare