Kling 4.0 Is Coming in October, and the Only Tier Shipping Today Is Kling 4.0 Flash

There are two products inside the name Kling 4.0, and today they sit in different states. Kling 4.0 is the full model: announced, specified on the vendor's own pages, and not released — the release note says only that it "will officially launch in October," without a day. Kling 4.0 Flash is the lighter tier, and it is the one that exists: early access opened on 28 September 2026 to a limited group, capped at 20 seconds and 720p. The generation everything in the vendor's own documentation is measured against is Kling 3.0, the flagship that has been shipping since February and the one our catalogue actually routes. So this piece is three things at once — a preview of a model that has been announced but not released, a reading of the only tier you can use right now, and a separation of the numbers the vendor has stated from the numbers it has only promised.

The reason to read it carefully rather than skim the summary is that the interesting material is in the qualifiers. Two of the headline capabilities on the marketing page are explicitly "coming soon" and not present in anything you can generate with today. The reference limits that read like round numbers on a slide are per-category caps inside a combined total, which changes what a real prompt can actually carry. And the panel at Advertising Week New York on 6 October, which is what the announcement post is mostly about, is not the launch — it is a talk scheduled ahead of one.

Two tiers, two states of existence

The vendor files Kling 4.0 and Kling 4.0 Flash as one model family with two tiers, and the split is clean. Flash generates up to 20 seconds in a single pass at up to 720p and is described as "built for frequent creation," available now in a limited rollout. The full Kling 4.0 generates up to 30 seconds in a single pass at up to 4K and is described as coming in October. Nothing in either description carries a price, and no rate card for either tier has been published.

• Access — Flash: limited early access from 28 September 2026, in the vendor's own app, for a named subscriber group. Full model: no public date beyond the month.

• Duration — Flash up to 20 seconds in one pass; full model 3 to 30 seconds.

• Resolution — Flash up to 720p; full model 720p, 1080p or 4K, with 21:9 ultrawide in the mix.

• Model identifier — none published for either tier, so there is nothing to put in a config file and no endpoint to point at.

• Price — none published for either tier.

If you are waiting for the full model, the honest position is that you are waiting on a month, not a date. If you want to use something now and you are not in the early-access group, the practical answer is the previous generation, which is a different product with a published price.

What the vendor says the full model will do

Everything below is from Kuaishou's own release note and feature page. None of it has been independently reproduced, and there is no third-party score for either tier — the tracker that ranks video models still carries Kling 3.0 rows and no Kling 4.0 row. Treat the list as a specification of intent, not as measured behaviour.

• One pass, 30 seconds — the full model generates up to 30 seconds in a single generation, which the vendor frames as room for "multiple story beats" inside one sequence rather than a clip assembled from stitched takes.

• Up to 10 keyframes — placeable anywhere on the timeline to steer camera movement, shot changes, character entrances and product reveals. This is the multi-shot control surface, and the count is the headline upgrade over the previous generation.

• Up to 15 references per task — and this is where the round number needs unpacking. The combined cap is 15 items across images, video, voice references and subjects, but the per-category limits are up to 10 images, up to 5 videos with a combined duration of up to 30 seconds, and up to 7 subjects of which up to 3 may be video-based. You cannot spend the whole allowance on images.

• Native 4K and 21:9 — with 10-bit HDR stated for both 1080p and 4K, which the feature page lists as coming soon rather than as present.

• Audio — two-channel stereo with tighter lip sync, dialogue across nine languages plus accents and dialects, and on-screen text generation in the same nine languages plus emoji.

• Prompt length — up to 8,000 tokens of input text.

• Re-editing — change an expression, a camera angle, a background or a single line of dialogue and the vendor says the rest of the frame stays intact.

• Video editing input — up to 5 input clips including one primary, each 3 to 30 seconds, with the combined duration capped at 30 seconds.

The three claims to hold loosely

The first is 10-bit HDR. The release note lists it as coming soon at both 1080p and 4K, and the feature page repeats the qualifier. That does not make it false; it makes it not yet a reason to choose the model, because a capability that is not in the shipped build cannot be tested on your footage.

The second is the resolution itself on the tier you can use. Flash tops out at 720p. The 4K conversation belongs to a model nobody outside the early-access group has generated with, and the two tiers are not the same product at different speeds.

The third is the "up to 15 references" figure. Read as one number it sounds like freedom; read as the per-category caps above it is a budget you allocate. The vendor is not hiding that — the limits are printed under the total in both the release note and the feature page — but the round number travels further than the footnotes, which is why it ends up in the summaries rather than the caps.

One thing that is verifiable is the vendor's own claim about how it was received by a practitioner. On 2 October the vendor published a hands-on account from Johnson Sheng, a creative partner at Kling AI and senior creative director at Ogilvy, describing access as a closed beta with an official launch still to come in October. He reports that a shot built from subject and scene assets alone — no storyboard — held up during a high-speed whip pan and a handheld push-in. That is a named working professional on the vendor's own site, which makes it more checkable than an anonymous benchmark table and still not independent: he is described as a creative partner of the vendor.

The Advertising Week panel is not the launch

The post that put this on the calendar says Kling AI will take the stage at Advertising Week New York on 6 October for a panel titled "The New Production Engine: Powering Creativity at Scale with Kling AI," and that ahead of the launch of Kling 4.0 the company will share "a closer look at how the next generation of AI video is shaping enterprise content production." Advertising Week New York 2026 runs 5 to 8 October in the Penn District in Manhattan.

Read the tense. "Ahead of the launch" is a company talking about a launch it has not held. A conference panel is a marketing event with a date on it; it is not a release, and the vendor's own release note still says October without a day. If you are planning around this, plan around the release note, not the agenda.

What this means if you are picking a video model this week

We do not route Kling 4.0 or Kling 4.0 Flash. We checked the catalogue today under every spelling of the vendor prefix and the tier name — the Kling family on our side runs from the 2.1 and 2.5 generations up through Kling 3.0 and its Omni variant, and stops there. There is no model identifier for the new tiers to add, so there is nothing to add.

What we do route is Kling 3.0, at $0.084 per request as our own catalogue lists it, on the same OpenAI-compatible key as every other model on the platform, with the provider's list price passed through and nothing added on top. For anyone whose decision this week is "what do I generate with while October plays out," that is a callable model with a published price and a shipped build behind it, reachable through one endpoint rather than a second vendor contract and a second SDK.

The second thing worth knowing is what happens after October. When the full Kling 4.0 opens, the useful pattern for a model this new is to try it behind failover rather than on a production path — route the request, let it fall through to a tier you have already validated if the new one errors or times out, and keep the client code unchanged. That is a configuration decision on our side, not a rewrite.

What to watch, and when

Three things carry the information from here. The 6 October panel is where the enterprise-workflow and pricing story will be told, if it is told at all. Sometime in October the full model opens, and the 10-bit HDR output the vendor currently marks "coming soon" is the one that will either arrive with it or slip again. And whenever a rate card appears, the comparison stops being about duration and keyframes and starts being about cost per finished shot.

One thing to check before you plan around either tier: the vendor has announced no API access for them. There is no model identifier and no endpoint, which means the only way anyone is calling Kling 4.0 today is inside the vendor's own creation page, and even that is limited to the Flash tier for a named subscriber group. An announcement with no identifier is a thing you can watch; it is not yet a thing you can build on.

The accurate summary is that Kling 4.0 is specified and not shipped, Kling 4.0 Flash is shipped to a limited group at 720p and 20 seconds, and the figures that sound best in the announcement are the ones the vendor has hedged hardest. That is not a reason to dismiss the model. It is a reason to keep the previous generation on the shortlist while the current one is still a promise with a month attached.