News / Model Watch · Video API + Business
Kling 3.0 API (2026): Official Pricing, Turbo & Omni — and the $3B Spin-Off, Explained for Builders
Kuaishou's Kling is having the loudest August of any Chinese video model that isn't named Seedance. The Kling 3.0 family — 3.0, 3.0 Turbo, and 3.0 Omni — is live on the official international API with per-second USD pricing from $0.084 (720p) to $0.42 (native 4K), the first native-4K direct output in the industry, and Q2 earnings that read like a fundraise deck: RMB 850M+ revenue, +200% YoY, 100M+ users. Behind it: a ~US$3B July financing at ~US$18B valuation and persistent spin-off-IPO reports. This brief verifies every number against the official pricing page, API docs, and Kuaishou's Q2 release — then tells you which variant to route.
Kling 3.0, 3.0 Turbo, and 3.0 Omni are available on Kuaishou's official international API today, per the developer pricing page and API docs (read 2026-08-21). List prices: $0.084/s at 720p, $0.112/s at 1080p, $0.42/s at native 4K (video Unit = $0.14); Turbo bundles native audio at $0.112–$0.14/s; Omni adds video input. The API is task-based REST with JWT auth (Access Key + Secret Key), not OpenAI-compatible chat. Kuaishou's Q2 2026 results (Aug 19) put Kling AI revenue above RMB 850M (+200% YoY) and confirm a July round of ~US$3B at ~US$18B post-money, Kuaishou retaining control.
From Sora challenger to spin-off candidate: the timeline
| When | What happened · source |
|---|---|
| Jun 2024 | Kling 1.0 launches — the first high-quality Chinese video model openly usable, positioned directly against OpenAI's Sora. |
| Dec 2025 | "全能灵感周" ships Kling Video 2.6 (audio-video together), the O1 unified multimodal video model, Image O1, and Avatar 2.0; monthly revenue crosses US$20M, ARR ~US$240M (per Huibo research aggregation). |
| Feb 2026 | Kling 3.0 series launches (official VIDEO 3.0 user guide published Feb 6): All-in-One full-modal input/output, up to 15s generation, multi-shot storytelling, synchronized audio-video. |
| Apr 2026 | Industry-first native 4K (3840×2160) direct output ships on the 3.0 series. |
| Q2 2026 | Kling 3.0 Turbo launches (CEO Cheng Yixiao: "第二季度我们推出了可灵3.0 Turbo模型") to cut creation cost; Kling MCP + CLI enable agent-driven batch creation. Two Kling-generated ads win at Cannes (1 silver, 2 bronze). |
| Jul 2026 | Beijing Kling subsidiary completes ~US$3B capital raise, ~US$18B post-money, Kuaishou retains control (confirmed in Q2 release); Tencent, Alibaba and Baidu reported among investors (HKET, Jul 2). |
| Aug 7 | CNMO reports core researcher Wang Xintao (Real-ESRGAN/GFPGAN author) has left Kling — unconfirmed by Kuaishou. |
| Aug 19 | Q2 results: Kling AI revenue >RMB 850M (+200% YoY); users >100M across 224 countries/regions; ~50K enterprise clients (both +67% vs end-2025). Chinese media widen IPO reporting. |
What's on the API, per official docs
- Three video tiers.
Kling 3.0(flagship: text/image-to-video, 3–15s, 720p/1080p/4K, optional native audio, custom start/end frames, multi-shot);Kling 3.0 Turbo(efficiency tier, native audio included, up to 1080p);Kling 3.0 Omni(adds video input on top of text/image — billed as its own variant). The docs overview pitches "Kling 3.0 & 3.0 Omni — synchronized audio-video generation, lower cost." - Native audio is a billing toggle, not a separate product. Same model, two rates: without native audio vs. with it (plus a voice-control add-on tier on standard 3.0).
- Task-based REST, not chat. Create a generation task, get a task ID, poll or webhook for the video URL. Auth is JWT (HS256) signed with an Access Key / Secret Key pair from the developer console — not an OpenAI-style bearer API key.
- Image side refreshed too. Kling Image 3.0, 3.0-omni, and O1 at $0.028/image (1K/2K; 4K $0.056); Image 2.1 remains at $0.014–$0.028. Avatar (digital human) at $0.056–$0.112/s and Motion Control at $0.126–$0.168/s round out the shelf.
Kling 3.0 API pricing (2026): official list, per second of generated video
Kling 3.0 starts at $0.084 per second (720p, no audio) and tops out at $0.42 per second at native 4K — 1 video Unit = $0.14 list price, read from the official pricing page on 2026-08-21.
| Model / mode | 720p | 1080p | 4K |
|---|---|---|---|
| Kling 3.0 — no native audio | $0.084/s | $0.112/s | $0.42/s |
| Kling 3.0 — with native audio | $0.126/s | $0.168/s | $0.42/s |
| Kling 3.0 Turbo — native audio included | $0.112/s | $0.14/s | — |
| Kling 3.0 Omni — text/image, no audio | $0.084/s | $0.112/s | $0.42/s |
| Kling 3.0 Omni — text/image + audio | $0.112/s | $0.14/s | $0.42/s |
| Kling 3.0 Omni — video input, no audio | $0.126/s | $0.168/s | $0.42/s |
| Kling Image 3.0 / 3.0-omni / O1 | $0.028/image (1K/2K) · $0.056 (4K) | ||
Source: official Kling developer pricing page, read 2026-08-21. Units are the platform's internal credit (video 1 Unit = $0.14, image 1 Unit = $0.0035 list); prepaid packages can discount below list. Third-party relays set their own prices — often higher than official list, occasionally promotional below it.
What a clip actually costs: 10-second math vs Seedance
A 10-second 1080p Kling 3.0 clip costs $1.12 without native audio ($1.68 with); Turbo does it with audio for $1.40; 4K runs $4.20. For context, our Seedance API guide carries relay-route estimates of ~$0.30–$0.40 per 10s for Seedance 2.0 — Kling 3.0 at official list runs 3–4× that, which is the price of native 4K, multi-shot control, and a two-year head start in video:
| 10-second clip (2026) | Official cost | Notes |
|---|---|---|
| Kling 3.0 · 720p · no audio | $0.84 | Cheapest official 3.0 path |
| Kling 3.0 · 1080p · no audio | $1.12 | Draft/iteration tier |
| Kling 3.0 Turbo · 1080p · audio | $1.40 | Q2 2026 cost-efficiency tier |
| Kling 3.0 · 4K · any mode | $4.20 | Native 4K direct output |
| Seedance 2.0 (relay est.) | ~$0.30–0.40 | Estimate per our Seedance guide, not Kling/official pricing |
Kling figures are exact multiplication of official per-second list prices; Seedance row is a labeled estimate. Compare quality before price — Artificial Analysis currently ranks both rivals ahead of Kling 3.0 Omni in text-to-video-with-audio (see below).
How to call it: JWT auth + create-task pattern (Python)
Unlike the OpenAI-compatible text APIs most of this site covers, Kling authenticates with a JWT you sign yourself from the Access Key / Secret Key pair issued at kling.ai/dev/api-key (secret shown once — store it immediately). Official docs' token recipe:
import time, jwt
ak = "YOUR_ACCESS_KEY"
sk = "YOUR_SECRET_KEY"
def encode_jwt_token(ak, sk):
headers = {"alg": "HS256", "typ": "JWT"}
payload = {
"iss": ak,
"exp": int(time.time()) + 1800, # valid 30 minutes
"nbf": int(time.time()) - 5,
}
return jwt.encode(payload, sk, headers=headers)
token = encode_jwt_token(ak, sk) # send as: Authorization: Bearer <token>
- Workflow: POST a generation task (text-to-video / image-to-video, model + duration + resolution + audio flags) → receive a
task_id→ poll the query endpoint or register a webhook until the video URL lands. Generation typically takes tens of seconds to minutes, not milliseconds — design for async. - Model names: use the exact model identifiers in the official API reference; third-party relay docs sometimes carry stale v1/v2-era names.
- Budget guard: bill per generated second means a retry loop multiplies cost fast — cap duration and retries per task in your own wrapper.
Which Kling model to route: 3.0 vs Turbo vs Omni
| Your workload | Route to | Why |
|---|---|---|
| High-volume social/ad content, audio wanted, cost-sensitive | Kling 3.0 Turbo | Native audio included at $0.112–$0.14/s — the tier Kuaishou built explicitly to lower creation cost (launched Q2 2026) |
| Multi-shot narrative, precise start/end frames, no audio needed | Kling 3.0 (no audio) | Cheapest per-second path to the flagship's cinematic control; add audio per clip when it matters |
| Editing/extending existing footage, video-conditioned generation | Kling 3.0 Omni (video input) | The only variant billing a video-input mode — full-modal in/out with synchronized audio-video |
| Final masters for film/brand work | Kling 3.0 · 4K | Industry-first native 4K direct output at $4.20/10s — skip the upscaler pipeline |
| Open-weight, self-hosted video generation | MiniMax H3 | Kling ships no weights; H3 is the open alternative — see our H3 open-weights brief |
The competitive picture: #3 on the audio-video leaderboard
- Kling 3.0 Omni ranks third in Artificial Analysis's text-to-video-with-audio category, behind ByteDance's Seedance 2.0/2.5 line and Alibaba's HappyHorse-1.0 (per iFeng coverage citing the leaderboard, Aug 2026). The same coverage counts nine of the global top ten text-to-video systems as Chinese.
- The founder effect is the storyline. HappyHorse is led by Zhang Di — the former Kuaishou VP who built Kling — after he moved to Alibaba. And on Aug 7, CNMO reported Wang Xintao (Real-ESRGAN/GFPGAN author, described as a 0-to-1 contributor) has left Kling; neither he nor Kuaishou has confirmed.
- Western rivals deprioritized the track. The Sora business wound down and Veo's iteration slowed (CNMO), while Chinese labs now supply global film/ad pipelines — Kling itself earns an estimated 70–75% of revenue overseas (Lan Dong Commerce, Aug 10), heavily North America.
The business story: RMB 850M quarter, $3B raise, IPO clock
- Q2 2026 (official, Aug 19): Kling AI revenue >RMB 850M (+200.0% YoY), dragging Kuaishou's "other services" segment +18.5% YoY. Global users >100M (+67% vs end-2025) across 224 countries/regions; enterprise clients ~50K (+67%).
- The July round is official — the IPO is not. Kuaishou's Q2 release confirms Beijing Kling signed a capital increase of up to ~US$3B at ~US$18B post-money, with Kuaishou keeping control and consolidating the unit. Spin-off-IPO framing comes from May's reported ~$20B Pre-IPO round (36Kr, iFeng) and August's IPO coverage — treat as reports until a filing exists.
- Why builders should care: a separately-capitalized Kling means an increasingly professionalized developer platform (the Q2 release already brags about MCP and CLI for agent batch creation), but also a standalone entity that can reprice Units, restructure plans, or split domestic/international stacks. Pin model IDs and prices; re-verify each quarter.
Get Chinese models through one OpenAI-compatible endpoint
ChinaModelAPI is an independent relay that brings Chinese frontier models — DeepSeek, Qwen, GLM, Kimi — to builders worldwide behind a single OpenAI-compatible API, with USDT/USD1 payments and no subscription. Kling's API is task-based video generation, a different shape from our chat-completions text routing — video-model relay support is on the roadmap, not live. Join the waitlist to hear first.
ChinaModelAPI is an independent relay with no official relationship to Kuaishou or Kling AI. All Kling prices above are official list prices from kling.ai, re-verified 2026-08-21.
Primary sources
- Kling developer pricing — official per-second Unit pricing for 3.0 / Turbo / Omni / Image (read 2026-08-21)
- Kling API docs — overview: "Kling 3.0 & 3.0 Omni synchronized audio-video generation"
- Kuaishou IR — 2026 Q2 and interim results (2026-08-19): Kling revenue >RMB 850M +200%, 3.0 Turbo, native 4K, MCP/CLI, ~US$3B raise
- 证券时报 — 快手第二季度收入355亿元 可灵AI收入增长超两倍 (2026-08-20)
- 北京日报 — 可灵AI营收同比增长超200%、全球用户破1亿、7月完成近30亿美元融资投后约180亿美元 (2026-08-19)
- 观察者网 — 快手财报"冰与火":可灵收入暴涨200%,3.0系列原生4K、MCP与CLI (2026-08-20)
- 香港經濟日報 — 快手次季業績:可靈AI收入逾8.5億增兩倍;北京可靈7月增資協議 (2026-08)
- 凤凰网 — 1359亿!快手可灵要IPO了:AA榜单 Kling 3.0 Omni 第三、张迪背景、4月原生4K (2026-08)
- 36氪 — 快手断臂求生,可灵出去找钱?分拆与 Pre-IPO 报道 (2026-08)
- CNMO/搜狐 — 可灵AI核心骨干王鑫涛离职报道 (2026-08-07)
- 腾讯新闻/蓝洞商业 — 可灵收入70–75%来自海外、年化冲击10亿美元 (2026-08-10)
- 新浪财经/WSJ — 中国AI视频崛起:AA前十中九个来自中国 (2026-08-20)
FAQ (2026)
How much does the Kling 3.0 API cost?
Official list (1 video Unit = $0.14): $0.084/s at 720p, $0.112/s at 1080p without audio; +~50% with native audio; $0.42/s at native 4K. Turbo bundles audio at $0.112–$0.14/s. A 10s 1080p clip: $1.12 no audio, $1.68 with.
Available outside China?
Yes — the international developer platform at kling.ai/dev bills in USD Units per generated second, plus many third-party relays resell Kling capacity at their own prices.
3.0 vs Turbo vs Omni?
Turbo = cost tier with audio included (Q2 2026). 3.0 = flagship multi-shot, optional audio, native 4K. Omni = adds video input with synchronized audio-video — the variant ranked #3 on Artificial Analysis.
Does it do 4K?
Yes — April 2026 brought the industry's first native 4K (3840×2160) direct output. On the API it's $0.42/s, so a 10s 4K master is $4.20.
Open weights?
No — API and subscription only. The open-weight Chinese video option is MiniMax H3 (HF + ComfyUI); see our H3 brief.
OpenAI-compatible?
No. Task-based REST with JWT auth (sign an Access Key/Secret Key pair yourself), create task → poll/webhook. Don't expect chat-completions shapes or bearer API keys.
What's the spin-off about?
Official: July's ~US$3B raise at ~US$18B post-money into the Beijing Kling subsidiary, Kuaishou retaining control. Reported, unconfirmed: a full spin-off IPO (May reports cited ~$20B Pre-IPO). Q2 revenue >RMB 850M, +200% YoY.
vs Seedance 2.0 / HappyHorse-1.0?
Kling 3.0 Omni is #3 on Artificial Analysis's text-to-video-with-audio board, behind both. Kling's edges: native 4K direct output, Turbo pricing tier, MCP/CLI agent tooling, 100M+ user base.