Frontline Hotspot
Frontline Hotspot

Kling 4.0 Lands in October: 30-Second 4K HDR vs Seedance

Kling 4.0 is official (September 28, 2026, per official announcements and media reports): full launch in October, with the lightweight 4.0 Flash opened the same day to black-gold annual members in a limited preview. Core specs: 4K and 1080p in 10-bit HDR, up to 30 seconds per generation, multi-round extension to 2 minutes, an 8,000-token prompt ceiling, up to 15 multimodal references (10 images + 5 videos + 7 subjects), and lip-synced multilingual output across Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French and Hindi; the creation page gains timeline-based continuous creation and a canvas Agent experience. Competitive context: a direct answer to Seedance 2.5 (launched July 31 to strong reception); Citi keeps Kuaishou at Neutral (target 41 HKD) with two questions - can Kling 4.0 pull Seedance users over, and is commercial pricing competitive - against a July funding round of about USD 3 billion at a USD 18 billion post-money valuation. Pricing and the exact launch date are unpublished; this piece withholds verdicts until real output lands.

Published October 3, 20269 min read
<!-- kling-4-0-official-launch-hotspot | hotspot | Kling 4.0 Lands in October: 30-Second 4K HDR vs Seedance -->

On September 28, 2026, Kling AI officially announced that Kling 4.0 will launch in October. On the same day, Kling 4.0 Flash, a lighter variant, opened a small-scale preview to Black Gold annual members. According to Kling's official announcement and media coverage, the core specifications are: 4K and 1080p output in 10-bit HDR format, single generations of up to 30 seconds, extension to 2 minutes through repeated continuation, and a prompt input limit raised to 8,000 tokens. Taken individually, none of these numbers is shocking. Stacked together, they describe a clear ambition: Kling intends to move the standard for a flagship video model from being able to generate footage to being able to deliver finished material.

The timing is telling. According to media reports, ByteDance's Seedance 2.5 has earned an exceptionally strong reputation since launching in late July and has become the first choice for a large number of creators. Alibaba's Wan 3.0 went live in late August, and MiniMax's video products keep pulling users in on price-performance. Announcing the full spec sheet at this exact moment is less a routine iteration than a direct challenge. This article breaks the story down: the specs worth examining one by one, what 4K HDR and 30 seconds actually mean for the industry, two perspectives on the competitive and capital picture, and what ordinary users can do before launch day. Every spec number below comes from official announcements as relayed by media; anything unannounced, we simply will not invent.

What Was Announced: The Spec Sheet, Line by Line

Kling's official positioning for version 4.0 reads as follows: major improvements in visual realism, creative controllability, and narrative completeness, with breakthroughs in handling large-scale scenes, audiovisual quality, professional-grade output specifications, multimodal references, precise video editing, and creative replication. Statements like these appear at every launch event, so the real information sits in the quantifiable specs. Here they are, one by one.

First, professional output. Kling 4.0 supports 4K and 1080p in 10-bit HDR. Ten-bit color depth offers several times the distinguishable levels of the common 8-bit, and combined with HDR's dynamic range, shadow and highlight transitions retain their detail. This is a specification you can hand into a professional post-production pipeline, not merely something good enough for a phone feed. AI-generated video has borrowed the word "cinematic" for two years now; with Kling 4.0, a leading Chinese model has, for the first time, written delivery-grade color specifications into its official parameter table.

Second, duration. A single generation runs up to 30 seconds, and repeated continuation can stretch a piece to 2 minutes. A 30-second single shot can carry a complete narrative arc rather than fragments stitched together, while the 2-minute ceiling leaves room for longer formats. For anyone producing short dramas, advertisements, or course content, these two numbers directly shape how storyboards get planned.

Third, prompt capacity. The input limit rises to 8,000 tokens, enough to hold a fairly detailed shooting script. The workflow of writing a script first and feeding it to the model becomes practical, and prompt engineering can shift from squeezing characters to writing clearly.

Fourth, multimodal references. A single generation can combine up to 10 images, 5 video clips, and 7 subjects, for a maximum of 15 items in one task. Images control style and scene, video references action and rhythm, and subjects preserve character consistency. Together they target the most painful point in commercial work: making AI output look like what you actually ordered.

Fifth, languages and lip sync. The official statement lists support for Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French, Hindi, and more, including regional accents, alongside lip matching, text generation, and multiple keyframes. For teams producing content for overseas markets, multilingual lip sync means one video can become several language versions at low cost.

Sixth, the creation interface: a unified input box, grid and list layouts, timeline-based continuous creation, and a new canvas Agent experience. This item is easy to overlook, but it determines daily usability. Timeline creation paired with 2-minute continuation suggests the product side was built with long-form content in mind.

Put the six items side by side and the product logic behind Kling 4.0 becomes fairly clear: this is not a single-record sprint but the completion of an entire production chain. Script writing (8,000 tokens), style control (15 references), footage generation (30-second single shots), length extension (2-minute continuation), and multilingual lip sync are each addressed, and the upgraded creation interface ties those stages into one timeline. In other words, what Kling wants to sell is not one dazzling generation but a pipeline you can run every working day. That is also its differentiation opportunity against Seedance 2.5: the rival has proven, through reputation, the strength of its single-point capability, while Kling's answer sheet is the completeness of the whole chain.

4K HDR and 30 Seconds: A Fight Over Standards

Placed against the wider industry, two aspects of this spec sheet deserve a closer look.

The first is that 4K 10-bit HDR is a competition over delivery specifications. For two years, video model comparisons centered on visual quality and motion plausibility, and a resolution figure in the spec table was enough. This time Kling wrote color depth and dynamic range into the official language, pushing the contest from "looks acceptable" toward "fits post-production, fits the big screen." For advertising and film professionals, this is the first time AI-generated footage could enter a professional color grading pipeline without conversion. The implication is much larger than simply "sharper pictures."

One common misreading deserves a caution. Some coverage frames 4K HDR as evidence that Kling simply outclasses its rival on picture quality. That conclusion comes too early. The two vendors publish different items in different formats in their spec tables, so comparing them line by line is not rigorous. Publishing delivery-grade specifications is itself worth crediting, but treating a published parameter as a victory over a competitor skips the step that matters: a real launch, with real footage, judged in real production.

The second is that 30 seconds is becoming the flagship baseline. According to Kling's official announcement and media coverage, Kling 4.0 generates up to 30 seconds per run. Seedance 2.5, per its official release materials, also supports 30-second native single shots with multi-turn extension, and its ultra-long mode on Jimeng reaches 3 minutes. Kling's 2-minute continuation looks less like catching up and more like leading vendors confirming the same route: generating one usable, complete shot at a time is now the floor of the next stage, not the ceiling. The multimodal reference matchup is equally direct. Seedance 2.5 accepts up to 50 reference items; Kling 4.0 takes 15, a smaller figure on paper, though Kling publishes a specific combination of 10 images, 5 videos, and 7 subjects, with a different emphasis. Which reference system works better in real production is a question for launch-day testing, and we will not prejudge it.

Facing Seedance: Product and Capital Perspectives

According to media reports, the backdrop to this announcement is impossible to separate from Seedance 2.5: released on July 31, it launched simultaneously on the professional tier of Doubao and on Jimeng, later opened to enterprise users through Volcano Engine, and has landed in industrial training, automotive design platforms, and embodied-intelligence training data. We covered its release and its impact on the video landscape in our Seedance 2.5 report and our analysis of the landscape shift. Whether Kling 4.0 can pull creators back is the most interesting suspense of the coming month.

The capital perspective is equally blunt. According to Cailian Press reporting, Citi maintained its "neutral" rating on Kuaishou with a target price of 41 Hong Kong dollars, listing two indicators to watch: whether Kling 4.0 can divert Seedance's user base, and whether its commercial pricing is competitive. Those two questions happen to be the product-level questions as well. The first awaits real migration data after launch; the second has no answer yet, because Kling has not announced pricing for version 4.0, and we will not fabricate one. For capital context: media reports state that Kling completed an independent funding round of roughly 3 billion US dollars in early July at a post-money valuation of about 18 billion, and that a national AI fund injected 1.4 billion yuan in late August. A war chest of that size says this battle is about far more than a product ranking.

One more detail belongs on the record. According to Shanghai Securities News, initial reactions to the announcement were described as "relatively muted." That is a reporter's observation, not a verdict on user sentiment; the full version is not yet open, and genuine feedback can only form after the October launch. Placed next to Citi's two questions, the market's stance on Kling 4.0 reads as: respect the specifications, withhold the judgment.

Whether users actually migrate comes down to three things: whether output quality genuinely pulls ahead, whether the workflow saves time, and what the price turns out to be. The first two require hands-on testing; the third remains open. It is also worth noting that Seedance 2.5's enterprise deployments are already running, and once creator habits settle in, switching costs only grow. For Kling, it is hard to overstate how critical the October window is.

Before Launch: Three Things You Can Do Now

First, check your access. Kling 4.0 Flash is open in a small-scale preview to Black Gold annual members; eligible users can check in the app whether they have received access. Everyone else should simply wait for the October full release. As of writing, the official word is only the month of October, with no more precise date.

Second, treat the 8,000-token prompt limit as homework. Before the specification goes live, the cheapest preparation is prompt practice: a structured, storyboard-style prompt, combined with the right way to feed 15 multimodal references, puts you a step ahead of most users on day one. A few practice directions are worth trying: break each shot description into fixed fields such as subject, action, camera movement, lighting, and atmosphere; test how to express character consistency through multi-subject references ahead of time; and organize continuous actions in chronological order so that a 30-second shot carries a complete narrative. None of this requires version 4.0. You can build the habit on the models available today. Our hands-on guide is already published: How to Use Kling AI: A Kling 4.0 Flash Starter Guide.

Third, compare the field. The 30-second club already has active members. Seedance 2.5 is live with real deployment cases. Alibaba's Wan 3.0 went live in late August, leaning on price-performance and document-to-video workflows (see our Wan 3.0 report). If you prefer data privacy with self-hosting, our companion piece covers LTX-2 going open. Once Kling 4.0 ships, we will update our four-way comparison (see the 30-second club comparison) with conclusions organized by use case, not by untested performance claims.

Closing: Wait for a Real Verdict

Anyone can run a spec announcement. Kling 4.0's real exam begins after the October launch: whether the pricing shows sincerity, whether the continuation workflow feels smooth in practice, and how stable those 15 references prove in real production. Two pieces of the puzzle are still blank: pricing is unannounced, and the exact launch date is unset. Those are the two details most worth watching over the coming month. Our position is straightforward: we do not pre-issue conclusions. When it launches, the output will speak for itself, and in the 30-second club, real footage is more honest than any press release.

Interaction

Will you wait for Kling 4.0, or have you already moved to Seedance 2.5? Tell us your choice and your reasons in the comments.

Sources

  • Shanghai Securities News (China Securities Net), September 29, 2026: Kling 4.0 announcement details and core specifications
  • Cailian Press, September 30, 2026: competitive backdrop and Citi rating commentary
  • Geek media report, September 28, 2026: Kling 4.0 Flash preview availability
  • Baidu Baike entry for Kling 4.0: cross-check of specification details
  • Note: the above are media relayings of official announcements; the body text consistently attributes them as "according to Kling's official announcement and media reports." Pricing and the exact launch date remain subject to Kling's official pages.

This article is AI-assisted and human-edited. Last updated: 2026-10-03

Related

Frontline Hotspot

OpenAI DevDay 2026: Dots moves in, Sol costs a fifth

OpenAI DevDay 2026 (September 29, 2026, official recap basis) packed more than 20 major announcements - 25 by item count - with ChatGPT at 1.2B weekly users. Headliner: Dots, an always-on agent that learns what matters to you and works on your behalf (available on Pro and Business Premium; Enterprise/Edu/Healthcare admins can opt into a beta, off by default). GPT-6.1 Sol delivers near-Astra intelligence at one-fifth of GPT-6 Astra's standard token prices (official pricing: $2 input / $0.10 cached / $10 output vs Astra's $10/$1/$50; cached input 50% cheaper than GPT-6 Sol). Official benchmarks worth citing: DeepSWE v1.1 parity with Astra at about one-fifth cost, Terminal-Bench Science at $5.47 per task vs Astra's $23.80, OSWorld 2.0 within 2.1 points of Astra at roughly one-seventh cost. Ultrafast hits 8x / 300 tokens-per-second in Codex and 6x in the API (Sol tier coming soon); Codex closes the cloud loop with reusable environments, a /agents multi-task CLI view, automated code review, and Security Cloud repo scanning that keeps running with your laptop closed; the Agents API gains Computer Use, and AWS partnership yields Bedrock Managed Agents powered by OpenAI.

Oct 2, 20269 min read
Frontline Hotspot

DeepSeek Ports the NVIDIA Stack to Ascend: Four Repos, One Day

On 2026-09-30 DeepSeek open-sourced its Huawei Ascend infrastructure components (official basis relayed via AI-tool aggregators): a full kernel stack mapped one-to-one to its NVIDIA-side repos - TileKernels (TileLang-based operator library) gaining an Ascend backend selected automatically at runtime so one Python API runs on both GPU and NPU, DeepGEMM-Ascend (fully API-compatible, BF16/FP8/FP4 GEMM and MegaMoE), DeepEP-Ascend (EPBuffer V2.5-compatible), and FlashMLA Ascend sparse-attention kernels (410 TFlops prefill at 95% of hardware peak, 360 TFlops decoding at 83%, with a deep-dive technical report, README official basis). "Multiple tests near hardware limits" is the official claim. Cold thought: DeepEP-Ascend ships no LICENSE file as of the snapshot, some fused kernels are CUDA-only, and the validated stack centers on Ascend 950 series.

Oct 1, 20269 min read
Frontline Hotspot

1.6T Params, 48B Awake: What Meituan LongCat-2.5 Hides

Meituan shipped LongCat-2.5-Preview on 2026-09-28 (official basis relayed via AI-tool aggregators): the 1.6T-total / ~48B-active MoE and 1M-token context carry over, max output is 128K tokens; the headline change is native image understanding merged into the base model (following the DiNA discrete-native-autoregressive route validated by LongCat-Next), plus long-horizon agent tasks across terminals, browsers, GUIs, spreadsheets and design tools. The API speaks both OpenAI and Anthropic protocols, so existing agent tools connect by swapping base_url, and sign-ups get 5M free tokens (relayed basis). Cold thought: no public benchmarks (against Opus 5.5 at $4/$20 per million tokens with public evals), and as of 2026-09-29 weights are unreleased with no 2.5 repo in the GitHub org - Preview-stage figures defer to official disclosure.

Sep 29, 20269 min read