Kling 4.0 Is Here: New Features, API Status & What’s Next

On October 14, a video team’s planning doc had a tidy line about Kling 4.0: “Flash for Ultra Yearly only, full model due in October, API unknown.”

Two days later, the middle part of that line was out of date. The team’s next job, a 30-second spot of a skincare bottle turning on a tabletop, now has a new model to try in the browser and the same open question on the developer side.

That split explains most of the Kling 4.0 release. Creators can use the model in the browser today, while the 4.0 route through the Kling AI API is still a promise with no date. Planning gets easier once those two sit in separate columns.

Update the Rollout Timeline First

The Kling 4.0 release date is no longer a target window. Here is the sequence from Kling’s and Kuaishou’s own announcements:

Kuaishou’s press release, issued in Hong Kong on October 16, says the model “officially launched on October 15.” Kling AI’s own launch release, published the same day, is more specific about access: Kling 4.0 is “fully available now on the Kling AI website,” and API access will come “soon.”

Neither release breaks access down by subscription plan, and neither says whether Flash has opened beyond Ultra Yearly users. If your team works across several accounts, open the creation page on the one you plan to bill and check what it offers before anyone books production time around it.

Sort the Kling 4.0 Features Into Live and Coming Soon

The October 16 launch materials describe these Kling 4.0 features as part of the release:

• Native single-take generation up to 30 seconds
• Up to 10 keyframe inputs
• Omni Reference with up to 15 references per task (up to 10 images, 5 videos and 7 subjects, per Kuaishou)
• Prompts up to 8,000 tokens
• Steadier handling of camera moves such as push-ins, pans, tracking, and orbit shots
• Access through Kling’s MCP server and CLI, so AI agents can use Omni Reference and keyframe-to-video

Kling’s September 30 announcement also listed two-channel stereo audio, improved lip sync, and wider language and accent support.

Other items are still written in the future tense. The October 16 releases list 10-bit HDR output at 4K and 1080p, a 21:9 ultrawide format, and video extension up to two minutes as coming soon. Kling and Adobe also say Kling 4.0 and Kling 4.0 Flash will arrive in Adobe Firefly, again without a date.

For the bottle spot, the live list covers most of the brief. Thirty seconds in one take removes a stitch point, and keyframes let the team pin the opening frame, the moment the label faces the camera, and the closing pack shot. HDR is the gap. If the spot is headed for a high-contrast display, note HDR as a later requirement and grade today’s output the usual way.

A long single take also gives the model more time to drift. Launch reels don’t show rejected takes, so whether the label stays readable for 30 seconds is something only a test with your own product images can answer.

Check Where the Kling AI API Stands

This part of the story has barely moved. Both October 16 releases say API access for enterprise customers and developers is coming soon. As of October 19, neither release gives a 4.0 model ID, request schema, rate limits, launch date, or API pricing. Check Kling’s developer platform yourself before anyone writes code against it.

MCP and CLI support is the new wrinkle. Agent workflows can reach Kling 4.0 today, which helps a creator automate their own projects. A production pipeline needs something else: a documented API with published limits and billing. Kling’s releases treat the two separately, and so should your roadmap.

WaveSpeed’s October 1 overview of Kling 4.0 release date and API availability was written before the full launch, so its rollout section is now dated. Its API advice holds up: don’t build on guessed model IDs or parameters. A model name copied from a third-party storefront or an older Kling doc can fail at the account boundary, and a cost estimate needs official rates for the route you will actually use.

Build the Pipeline on Current Models Now

Waiting for the API doesn’t have to stall production. Kuaishou Kling has shipped two major versions this year, Kling 3.0 in February and 4.0 in October, and at the current pace of AI video model 2026 releases, the next switch is never far off. A repeatable test pack absorbs that churn, and the bottle spot can go into one today:

• One saved prompt covering the action, camera path, framing, and sound
• Product images and reference clips with documented usage rights
• A fixed duration and aspect ratio
• An acceptance sheet for label legibility, bottle shape, motion, and audio

Run the pack on the model your pipeline already calls through an API. Log the model name, account, settings, failed takes, reruns, review time, and cost per approved clip. Then run the same pack on Kling 4.0 in the web interface and label it as a manual test. That gives you a quality comparison now, while keeping the browser result apart from any claim about API throughput or price.

Keep the integration layer thin. If the code that submits jobs sits behind one adapter, adding a Kling 4.0 route later means writing one new adapter instead of rewiring the pipeline. Leave its parameters blank until Kling publishes them.

Switch When the API Is Documented

Set the switch rule now, so nobody has to debate it on the day the announcement lands. A workable version has four conditions:

1. Kling publishes a dated 4.0 model entry with a request and response format.
2. Your account can make an authorized call and get a result back.
3. Published pricing and limits fit the batch you plan to run.
4. The 4.0 output clears the same acceptance sheet as your current model.

When all four hold, move one bounded job, such as the bottle spot, and compare its log with the old route before moving anything else. Until then, keep shipping on the pipeline that works, and recheck Kling’s announcements with the date written next to each one.

Related: Best AI Cartoon Video Models in 2026: Kling vs Seedance vs MiniMax vs Wan