Kling 4.0 is the new generation of Kuaishou's video model. It generates up to 30 seconds of video in a single pass — twice as much as Kling 3.0 — and lets you steer the story with 10 keyframes. The model was announced on September 28, and the full version ships in October. For now only the lightweight Kling 4.0 Flash is available, in a closed beta for Kling AI annual subscribers. The current generation, Kling v3, already works in GPTunneL.
What's new in Kling 4.0
- 30 seconds per pass. Clips from 3 to 30 seconds with no stitching. Kling 3.0 topped out at 15 seconds.
- Extension to 2 minutes. A finished clip can be grown step by step while keeping characters and setting. Marked "coming soon".
- 10 keyframes. Previously there were two anchors — the first and last frame. Now you can place up to 10 reference images, and the model builds a continuous video between them.
- 15 references per task. Omni Reference mode accepts up to 10 images, up to 5 videos with a total length of up to 30 seconds, and up to 7 saved characters or objects.
- Character voice. A voice reference can be attached to a saved character — it will speak in its own voice in any scene.
- Stereo audio. Sound is generated together with the video, and lip sync is more accurate.
- 4K and 10-bit HDR. Resolutions of 720p, 1080p and 4K. HDR for 1080p and 4K is marked "coming soon".
- 21:9 format. An ultra-wide cinematic format joins 16:9, 1:1 and 9:16.
Official demos
All clips below are Kling 4.0 demos from the official Kling AI website.
30 seconds in one clip: a jump from a plane and a wingsuit flight over the desert in 21:9. You can see how the model handles fast motion and several people in frame:
A pianist on a ship's deck in a storm. Shots change from wide to close-up, and the character stays the same:
A fight in a hotel lobby — fast choreography of several characters, where video models usually mix up arms and bodies:
A swordswoman on a mountain path — detail in fabric, face and fog:
Kling 4.0 Flash vs the full version
Right now only Flash can be run. According to early reviews, it is noticeably simpler than the full version:
| Kling 4.0 Flash | Kling 4.0 | |
|---|---|---|
| Access | since September 28, beta for annual subscribers | October 2026 |
| Duration | 3–20 seconds | 3–30 seconds |
| Resolution | 720p | 720p, 1080p, 4K |
| Color | 8-bit SDR | 10-bit HDR (coming soon) |
Flash is the fast option for drafts. Kling has not announced pricing for either version yet, and the developer API is not open either.
How Kling 4.0 differs from Kling 3.0
Kling 3.0 came out on February 4, 2026: up to 15 seconds, native 4K at 60 fps, audio generated with the video and a multi-shot mode. In 4.0 the length doubled, and up to 8 more anchor points joined the first and last frame. Voiced characters, stereo audio, a 21:9 format and HDR were added.
The main competitor is ByteDance's Seedance 2.5: also up to 30 seconds per pass and up to 50 references. Kling 4.0 has fewer references but offers keyframes and voiced characters.
How to try Kling in GPTunneL
We will add Kling 4.0 to GPTunneL once Kuaishou opens the API. Meanwhile you can work with Kling v3 — prompts written for it will carry over to the new version:
- Open Kling v3 in the lab.
- Describe the scene in text or upload a starting image.
- Choose resolution and duration: $0.34 per second in 720p, $0.42 in 1080p.
- Run the generation and download the finished clip.
For drafts, Kling v3 Motion is cheaper — from $0.11 per second. No subscription needed: you pay per second of finished video. More on prompts in our Kling AI guide.
Kling 4.0 FAQ
When will Kling 4.0 be released? The full version is promised for October 2026. Kling 4.0 Flash has been running since September 28 in a closed beta for annual subscribers.
Is Kling 4.0 free? No. Flash is available only on a paid annual subscription, and pricing for the full version has not been announced.
Can I download Kling 4.0? No, the model runs only in the cloud. Sites offering to "download Kling for free" are not Kling.
Will Kling 4.0 be available in GPTunneL? Yes, once the API opens. The news will be in the blog and our Telegram channel.