The 9 longest AI video clips, ranked from 8 to 30 seconds

The longest single clip any AI video model will produce, according to its own maker, is 30 seconds, and three companies now state that ceiling. As of 14 October 2026 the documented maximum clip length across the field runs from 8 seconds at Google to 30 seconds at Alibaba Cloud, Kuaishou and ByteDance.

This list ranks nine models on one number: the maximum clip length the maker's own documentation states, read off the maker's page on 14 October 2026. No hands-on tests were run for it, and no figure comes from a comparison blog. Where two models state the same maximum, the one that is generally available ranks above one that is still rolling out.

One candidate did not make the cut. Higgsfield states no maximum clip length on its own site for a video made with GPT-6 Astra; the "over 10 minutes" figure that circulates comes from an OpenAI customer story, a marketing testimonial rather than a published spec, so it is left out. Google also documents an extend workflow for Veo 3.1, but the ranking uses the per-generation figure Google publishes. In all, nine models made the cut.

1. Wan 3.0 (Alibaba Cloud)

Wan 3.0 is Alibaba Cloud's All-in-One video model, and Alibaba's own documentation gives the ceiling: the model "supports up to 30 seconds per generation" at 30 frames per second. That is the longest figure in this list, tied with Kling 4.0 and Seedance 2.5, and Wan 3.0 is the one of the three that Alibaba sells through its public cloud Model Studio rather than an invitation.

The same pass generates dialogue, background music and sound effects, so a 30-second clip arrives with audio attached. Alibaba's guide also lists an extension feature and a smart-duration mode that picks a length from the prompt. For a creator who needs one unbroken take instead of a stitched sequence, Wan 3.0 is the safest of the three 30-second models to plan around.

Card

• Model: Wan 3.0
• Maker: Alibaba Cloud
• Max clip length: 30 seconds in a single generation
• Native audio: Yes, dialogue, music and sound effects
• Where to run it: Alibaba Cloud Model Studio

2. Seedance 2.5 (ByteDance)

Seedance 2.5 is ByteDance's audio-video model, and the company's Seed page describes it as "built for 30-second storytelling." ByteDance states that creators can make "videos up to 30 seconds in a single generation," with the option to extend twice. That matches Wan 3.0 and Kling 4.0 for the longest clip on the list.

ByteDance frames the model around longer narratives and reference control, saying motion is smoother and that Seedance 2.5 reads reference videos more precisely to go beyond motion transfer. The 30-second generation is the headline spec. It ranks a step above Kling 4.0 here only on availability, because Kling's full 30-second model is still rolling out.

Card

• Model: Seedance 2.5
• Maker: ByteDance
• Max clip length: 30 seconds in a single generation, extendable twice
• Native audio: Yes, audio and video are generated jointly
• Where to run it: ByteDance's Seed platform and API

3. Kling 4.0 (Kuaishou)

Kling 4.0 is Kuaishou's next video model, and its own FAQ gives the duration: "Kling 4.0 supports native video generation from 3 to 30 seconds," with up to 10 flexible keyframes. The top of that range ties it with Wan 3.0 and Seedance 2.5.

The catch is availability. Kling's page says the full model "is scheduled to launch in October 2026," with early access to the new creation experience rolling out from 28 September. Only a Flash variant is live, so for most users the 30-second figure is a promise rather than a setting today. Kling 4.0 adds two-channel stereo audio and tighter lip-sync, which keeps it in the top three despite the wait.

Card

• Model: Kling 4.0
• Maker: Kuaishou
• Max clip length: 30 seconds (range 3 to 30 seconds)
• Native audio: Yes, with two-channel stereo
• Where to run it: kling.ai (early access; full model promised for October 2026)

4. Luma Ray3.2 (Luma AI)

Luma's Ray3.2 is the longest model here built for editing footage a creator already has. Luma's own Ray3.2 product page states that Modify Video V2 runs "Up to 20 seconds, all at 1080p," and the model's release note repeats the figure for clips "up to 20 seconds long at 1080p." That sits between the 30-second leaders and the 10-second group.

Version matters here. Luma's newer Ray3.14 build, the current one on the same product page, lists 5 or 10 second text-to-video generations and Modify Video input up to 18 seconds, with extension reaching about 30 seconds. The 20-second figure belongs to Ray3.2, not to that newer build. For creators feeding their own footage in, Ray3.2's stated window is still the wider one.

Card

• Model: Luma Ray3.2
• Maker: Luma AI
• Max clip length: 20 seconds at 1080p (Modify Video)
• Where to run it: Luma Dream Machine and the Ray3.2 API

5. MiniMax H3

MiniMax H3 is MiniMax's general-purpose multimodal model, and the company's launch post states the number: video "with native stereo sound, up to 15 seconds at 2K resolution." That puts H3 between Luma's 20 seconds and the 10-second group.

MiniMax positions H3 as an omni-modal model that reads text, images, video and audio in a single context, and it runs at hailuoai.video. Fifteen seconds is half the field's ceiling, but the model generates its own stereo audio and defaults to 2K, so it trades raw length for resolution and sound. It is the newest MiniMax model on this list by a wide margin.

Card

• Model: MiniMax H3
• Maker: MiniMax
• Max clip length: 15 seconds
• Native audio: Yes, native stereo
• Where to run it: hailuoai.video

6. Runway Gen-4.5

Runway Gen-4.5 is the only model here that lets a creator pick any whole-second length inside its range. Runway's own help page lists "Supported durations: 2 - 10 seconds," so the maximum is 10 seconds and the minimum is 2.

Gen-4.5 is Runway's general model for text-to-video and image-to-video, running at 720p on the web for Standard plans and above. It is not the longest clip on this list, but the 2-second floor is useful for cheap motion tests before committing to a full run, and the same 12-credit-per-second rate applies across the range.

Card

• Model: Runway Gen-4.5
• Maker: Runway
• Max clip length: 10 seconds (range 2 to 10 seconds)
• Where to run it: Runway web app, Standard plan and higher

7. Pika 2.5

Pika 2.5 is Pika's flagship video model, and the company's own app page answers the length question: "Five seconds from a prompt alone. Attach a reference image and you can pick five or ten." So the maximum is 10 seconds, reached only through image-to-video.

Pika 2.5 is tuned for 1080p and for animating stills, including ordered keyframes that produce a one-to-five second transition between each pair. It matches Runway Gen-4.5 at 10 seconds, but a text-only prompt caps at 5 seconds, so a clip built from words alone lands shorter than Gen-4.5's.

Card

• Model: Pika 2.5
• Maker: Pika
• Max clip length: 10 seconds with an image (5 seconds from text)
• Where to run it: Pika

8. Hailuo 2.3 (MiniMax)

Hailuo 2.3 is MiniMax's earlier consumer model, still served on the Hailuo AI site, and its model page lists a "10s Max Clip Length" with 6-second and 10-second clips at 768p or 1080p. It ties Pika 2.5 and Runway Gen-4.5 at 10 seconds.

MiniMax has since moved its frontier work to H3, which reaches 15 seconds, so Hailuo 2.3 ranks below it. The model stays the cheaper, faster option on Hailuo AI for short social clips, and it is the older of the two MiniMax models in this list.

Card

• Model: Hailuo 2.3
• Maker: MiniMax
• Max clip length: 10 seconds
• Where to run it: hailuoai.video

9. Veo 3.1 (Google)

Veo 3.1 is the shortest on the list by its maker's own numbers. Google's Veo 3.1 documentation lists the video lengths: "4, 6, or 8 seconds," so 8 seconds is the per-generation maximum. Google's Gemini API reference describes Veo 3.1 as a model for "generating 8-second videos (720p, 1080p, or 4k) with natively generated audio."

Google's documentation also lists an extend feature that chains generations into a longer video, but Google states no single maximum for the result, so the ranking uses the documented per-clip figure of 8 seconds. A creator building a longer sequence here plans around several short shots and an editor, where the three 30-second models at the top of this list are built to hold one take together.

Card

• Model: Veo 3.1
• Maker: Google
• Max clip length: 8 seconds per generation
• Native audio: Yes, natively generated
• Where to run it: Gemini API and Google Cloud's agent platform