Best AI Video Generators in 2026: 20 Video Models Ranked

The best AI video generators in November 2026 are ByteDance's Seedance 2.5, MiniMax H3 and Alibaba's Wan 3.0, the three models that place highest on average across the four public video leaderboards. Google's Gemini Omni, which leads LMArena's text-to-video board, comes fourth, followed by Black Forest Labs' FLUX 3 and xAI's Grok Imagine. Every model in the top 10 generates sound with the picture, and the best make clips of up to 30 seconds in one go.

This guide ranks the top 10 video generation models on four arenas where people vote blind between clips, compares 20 models on length, resolution, audio and price, and covers text-to-video, image-to-video and the best open-source models. Every score, price and date was checked on November 2, 2026. One big name is missing: OpenAI shut down Sora in 2026. For still images, see our ranking of the best AI image generators.

Top 10 AI video generation models in 2026, ranked

Ranked by average position on four leaderboards: LMArena and Artificial Analysis, each for text-to-video and image-to-video. Only models you can use today count; Meta's Muse Video is previewed but not released.

1. Seedance 2.5 (ByteDance): first on Artificial Analysis' image-to-video board and in the top seven on all four. Makes 30-second clips with audio from up to 30 reference images.

2. MiniMax H3 (MiniMax): first on LMArena's image-to-video board and the best model with open weights. Several other top models are built on it.

3. Wan 3.0 (Alibaba): first on Artificial Analysis' text-to-video board and on both arenas' video-editing boards. Makes 30-second clips with dialogue, music and sound effects.

4. Gemini Omni 1.1 Flash (Google): first on LMArena's text-to-video board and second for image-to-video. It replaced Veo in the Gemini API and is free to try in YouTube Shorts.

5. FLUX 3 Video (Black Forest Labs): third on LMArena's text-to-video board (preliminary), with 20-second clips.

6. Grok Imagine Video 1.5 (xAI): fourth on LMArena's text-to-video board (preliminary), with sound and 1080p, from $0.02 per second.

7. HappyHorse 1.1 (Alibaba): Alibaba's second video line. Version 1.0 topped Artificial Analysis anonymously in April 2026 before Alibaba revealed it.

8. Veo 3.1 (Google): Google's previous flagship, still in the Gemini app and Flow, with native audio and 4K output on Google Cloud.

9. Kling 4.0 (Kuaishou): released October 15, 2026, with 30-second clips in 4K HDR. Too new to be rated; Kling 3.0 places 17th to 19th.

10. Vidu Q4 (ShengShu): a preview since October 7, 2026, with 4K 10-bit output and consistent character voices. Rated on its predecessor, Vidu Q3.

How the top 10 placed on each board

Position on each leaderboard. Where a board hasn't rated a model's newest version yet, the previous version counts; a dash means the model isn't on that board.

Seedance 2.5 and MiniMax H3 tie on average; Seedance places higher on the more recent Artificial Analysis boards.

Key numbers

• No single winner. Each of the top four leads one board: Gemini Omni and MiniMax H3 on LMArena, Wan 3.0 and Seedance 2.5 on Artificial Analysis.

• China leads. Six of the top 10 come from Chinese companies: ByteDance, MiniMax, Alibaba (twice), Kuaishou and ShengShu.

• Sound is standard. All 10 generate audio with the video: dialogue, sound effects and music. Among the 20 models here, Luma Ray3.2, Midjourney, HunyuanVideo and Adobe's Firefly Video Model don't.

• Clips got longer. Seedance 2.5, Wan 3.0 and Kling 4.0 make 30 seconds in one generation. Veo 3.1 makes 8 seconds at a time.

• One open model under many. Six of the top 10 on Artificial Analysis' image-to-video board are MiniMax H3 or built on it: H3 itself, Utopai X, Hyperflow H3, HeyGen Video 1, Pruna's P-Video-2-Pro and fal's H3 Max.

• Prices vary 20-fold. At around 720p, API prices run from $0.02 per second (Pika H3 Flex) to $0.40 (Veo 3.1 with audio). A 10-second 720p clip costs about $1 with Wan 3.0 or Gemini Omni and $4 with Veo 3.1.

• Sora is gone. OpenAI closed the Sora app in April 2026 and its API in September.

AI video arena leaderboards

Video arenas show voters two clips made from the same prompt by anonymous models and ask which is better. Thousands of votes become an Elo-style score: a higher score means the model wins more head-to-head comparisons. LMArena (now at arena.ai) and Artificial Analysis run the two main video arenas, each with a text-to-video and an image-to-video board. Artificial Analysis also splits its boards into clips with and without audio. Scores from different arenas aren't comparable, and new entries with few votes are preliminary.

LMArena text-to-video

Top 10 on the leaderboard dated September 21, 2026.

LMArena image-to-video

Top 10 on the leaderboard dated September 21, 2026.

Artificial Analysis text-to-video (with audio)

Top 10 on the Text to Video arena, viewed November 2, 2026.

Without audio, HeyGen Video 1 leads (1143), ahead of MiniMax H3 Max and FLUX 3 Max.

Artificial Analysis image-to-video (with audio)

Top 10 on the Image to Video arena, viewed November 2, 2026.

For editing existing video, Wan 3.0 leads both LMArena's video-edit board and Artificial Analysis' editing board with audio, with Seedance 2.5 second on both; Gemini Omni Flash 1.1 leads editing without audio.

The 20 best video models compared

Clip length is the most one generation makes. API prices are per second of video from each official pricing page, checked November 2, 2026, at 720p unless noted.

The top 10 video models in detail

1. Seedance 2.5

Seedance 2.5 is ByteDance's newest video model, launched July 31, 2026: up to 30 seconds per generation with joint audio, and up to 30 reference images and 10 video clips in one pass. It is first on Artificial Analysis' image-to-video board and in the top seven on all four. Its predecessor, Seedance 2.0, drew copyright complaints from the Motion Picture Association, Disney and Paramount in February 2026.

• API price: token-based on BytePlus, about $0.23 per second at 720p (BytePlus)

• Use it in: Dreamina and CapCut for subscribers (rolling out outside the US since late October), and Jimeng and Doubao in China

2. MiniMax H3

MiniMax H3 is the model behind MiniMax's Hailuo app, launched July 31, 2026, with 15-second clips, native stereo audio and 2K output. MiniMax released its base weights three days later, and they have become a foundation for others: HeyGen Video 1, Utopai X, Pika's H3 Flex and fal's H3 Max are all built on H3. The open license excludes the US, EU, UK and South Korea and requires authorization for products with more than $20 million in yearly revenue.

• API price: $0.08 per second (768p), $0.13 (2K); faster tiers from $0.025

• Use it in: Hailuo AI and the MiniMax API

3. Wan 3.0

Wan 3.0 is Alibaba's Tongyi Lab video model, generally available since August 24, 2026. It makes up to 30 seconds per generation, twice Wan 2.7, with dialogue, background music and sound effects, and can take documents and web pages as references. It is first on Artificial Analysis' text-to-video board and on both arenas' editing boards. Wan 3.0 is API-only; the last open Wan weights are Wan 2.2 (Apache 2.0).

• API price: $0.05 (480p), $0.10 (720p), $0.20 (1080p) per second (Alibaba Cloud)

• Free: 30 seconds of free quota in Alibaba Cloud Model Studio

4. Gemini Omni 1.1 Flash

Gemini Omni 1.1 Flash is from Google DeepMind's new model family, introduced at Google I/O in May 2026 as a model that can create from any input, starting with video. Version 1.1 Flash went generally available on August 27, 2026, with scene extension up to 40 seconds and first- and last-frame control, and Google shut down the Veo previews in the Gemini API in October.

• API price: about $0.10 per second at 720p, per Google

• Free: in YouTube Shorts Remix and YouTube Create, and with Google Flow's 50 free daily credits; the Gemini app needs a Google AI plan

5. FLUX 3 Video

FLUX 3 Video is part of FLUX 3, Black Forest Labs' first family that generates video with native audio, announced July 23, 2026 and on its API since August 4. It makes up to 20 seconds per generation in 720p, with upscaling to 4K. The German lab has announced an open-weight FLUX 3 Dev with no date.

• API price: $0.15 per second (720p) to $0.41 (4K), pay as you go

6. Grok Imagine Video 1.5

Grok Imagine Video 1.5 is xAI's video model, generally available since June 16, 2026, generating sound effects, ambience and dialogue in the same pass as the video. It makes up to 15 seconds at 1080p and accepts up to 14 reference images. SpaceX acquired xAI in a deal announced in February 2026.

• API price: $0.08 per second; Lite $0.02 (xAI)

7. HappyHorse 1.1

HappyHorse 1.1 is a closed video model from Alibaba's Token Hub unit, separate from the Wan team. Version 1.0 topped Artificial Analysis' text-to-video board in April 2026 before anyone knew who made it; Bloomberg revealed Alibaba as the developer. Version 1.1 followed on June 22, 2026, with lip-synced dialogue; a separate HappyHorse model edits existing video.

• API price: $0.07 (480p) to $0.18 (1080p) per second

• Use it in: happyhorse.com, the Qwen app and Alibaba Cloud

8. Veo 3.1

Veo 3.1, from October 2025, is Google's previous flagship, still in the Gemini app, Google Flow and Google Cloud. It generates dialogue, sound effects and ambient sound, with 4K output on Google Cloud. In the Gemini API, Google has replaced it with Gemini Omni.

• API price: $0.40 per second with audio, $0.20 without, on Google Cloud

• Free: through Google Flow's 50 free daily credits

9. Kling 4.0

Kling 4.0, from Kuaishou's Kling unit, launched October 15, 2026 with 30-second single-pass clips, up to 15 references and 4K output with HDR. It isn't on the leaderboards yet; Kling 3.0 places 17th to 19th. Bloomberg reported in October that Kling, now spun out of Kuaishou, picked banks for a Hong Kong IPO that could raise more than $1 billion. Kling 3.0 is also offered inside Adobe Firefly.

• API: Kling 4.0 API coming soon; Kling 3.0 is available, for example through fal at $0.168 per second with audio

10. Vidu Q4

Vidu Q4 Preview is ShengShu Technology's flagship, released October 7, 2026, with 4K 10-bit color and up to three voice references for consistent character voices. It isn't rated yet; its predecessor, Vidu Q3 Pro, ranks 18th on LMArena's image-to-video board.

• API price: from $0.045 per second at 540p, per ShengShu's launch release

10 more video models worth knowing

Runway Gen-4.5 and Aleph

Runway Gen-4.5 and Aleph come from Runway, one of the first companies to sell AI video generation. Gen-4.5 (December 2025) is its newest generation model and Aleph 2.0 edits existing video from instructions. Runway raised $315 million at a $5.3 billion valuation in February 2026 and is now building world models; its own models rank below the leaders on the arenas.

• Price: free 125 credits; Standard $15 a month; API $0.12 per second (Gen-4.5)

Luma Ray3.2

Luma Ray3.2 (June 2026) is Luma AI's video model with keyframe control and native HDR and EXR output for professional post-production. It doesn't generate audio. Luma raised a $900 million Series C led by HUMAIN in November 2025.

• Price: Plus $30 a month; API $0.30 per 5-second 720p clip

LTX-2.5

LTX-2.5 (August 2026), from Israel's Lightricks, is an open-weight video model with audio and 4K output. Its weights are on Hugging Face under the LTX community license, free for companies with less than $10 million in yearly revenue. Lightricks plans to split into two companies, Facetune and LTX.

• Price: free 800 credits; Lite $15 a month; API from $0.09 per second (720p)

PixVerse V6

PixVerse V6 (March 2026) makes 15-second clips with audio from the Singapore-based PixVerse, which reports more than 150 million registered users and raised $439 million in 2026.

• Price: API from $0.04 per second at 720p without audio

Pika H3 Flex

Pika H3 Flex (October 29, 2026) is Pika's open set of adapters post-trained on MiniMax H3, making video with synchronized audio in six sampling steps. Pika's own Pika 2.5 model is still available.

• Price: Starter $10 a month; API from $0.01 per second (480p)

Midjourney V1 Video

Midjourney V1 Video (June 2025) animates images into 5-second clips that can be extended, at 480p by default. It is image-to-video only, has no audio and no API, and hasn't been updated since.

• Price: included in Midjourney plans from $10 a month; a video job costs about 8 times an image job

Moonvalley Marey

Moonvalley Marey is trained only on licensed footage and aimed at film and advertising production, with camera and motion controls. In June 2026 the AI lab Reka combined with Moonvalley.

• Price: Standard $14.99 a month; no free plan

HunyuanVideo 1.5

HunyuanVideo 1.5 (November 2025) is Tencent's lightweight 8.3B open-weight model for text- and image-to-video at up to 720p, without native audio. Its license excludes the EU, UK and South Korea.

• Price: free to download and self-host

Adobe Firefly Video Model

Adobe Firefly Video Model makes 5-second 1080p clips trained on content Adobe has rights to, without native audio. For more, Adobe leans on partners: the Firefly app offers more than 30 models, including Veo 3.1, Kling 3.0 and Runway Gen-4.5.

• Price: included in Firefly plans

HeyGen Video 1

HeyGen Video 1 (September 30, 2026) is the avatar company's general-purpose video model, built on MiniMax H3: 5- to 15-second clips at up to 2K with stereo sound. It leads Artificial Analysis' text-to-video board without audio and is sixth with audio.

• Price: API $0.03 per second at 768p with audio; in the HeyGen app and through OpenRouter and other hosts

Best open-source video generation models

• MiniMax H3: the top-ranked model with open weights (33B base checkpoints), but its license excludes the US, EU, UK and South Korea.

• LTX-2.5: open weights with audio and 4K output, free for companies under $10 million in yearly revenue.

• Wan 2.2: Alibaba's last open Wan release (Apache 2.0, July 2025).

• HunyuanVideo 1.5: Tencent's lightweight 8.3B model, without audio.

• Pika H3 Flex: open adapters on top of MiniMax H3, under the same license.

• Krea Realtime 14B: Apache 2.0, built on Wan 2.1, for real-time streaming video.

• Mochi 1: Genmo's Apache 2.0 model from October 2024, now well behind the leaders.

Best AI video generator by use case

• Best overall: Seedance 2.5 and MiniMax H3, with Wan 3.0 close behind.

• Best text-to-video: Gemini Omni 1.1 Flash on LMArena, Wan 3.0 on Artificial Analysis.

• Best image-to-video: MiniMax H3 on LMArena, Seedance 2.5 on Artificial Analysis.

• Longest clips: Seedance 2.5, Wan 3.0 and Kling 4.0, at 30 seconds per generation.

• Best for 4K: Kling 4.0 (HDR), Vidu Q4 (10-bit), Veo 3.1 on Google Cloud and LTX-2.5.

• Best for editing video: Wan 3.0 and Seedance 2.5 lead the editing boards; Runway Aleph is built for it.

• Best free: Gemini Omni in YouTube Shorts and Google Flow's free daily credits; Wan 3.0's free API quota; Runway's and LTX's free starter credits.

• Cheapest API: Pika H3 Flex from $0.01 per second and Grok Imagine Video Lite at $0.02.

• Safest for commercial work: Adobe Firefly and Moonvalley Marey, trained on licensed content.

• Talking avatars: a different category; avatar tools such as HeyGen and Synthesia make presenter-style videos from a script.

Where to use these models

Most people meet these models inside apps: CapCut and Dreamina (Seedance), Hailuo (MiniMax H3), Google's Gemini app, Flow and YouTube (Gemini Omni and Veo), Grok (Grok Imagine) and the Kling app. Creative platforms such as Higgsfield, Krea and Adobe Firefly put many models behind one interface, and developers often use fal, which hosts many of the models here behind one API.

What changed in 2026

FAQ

What is the best AI video generator in 2026? ByteDance's Seedance 2.5 and MiniMax H3 place highest on average across the four public video leaderboards, just ahead of Alibaba's Wan 3.0. Google's Gemini Omni leads LMArena's text-to-video board.

What is the best free AI video generator? Google's Gemini Omni is free in YouTube Shorts Remix and the YouTube Create app, and Google Flow gives everyone 50 free credits a day. Alibaba's API gives 30 free seconds of Wan 3.0, and open models such as LTX-2.5 and HunyuanVideo are free to run on your own hardware.

Which AI video generators make sound? All of the top 10: Seedance 2.5, MiniMax H3, Wan 3.0, Gemini Omni, FLUX 3, Grok Imagine, HappyHorse, Veo 3.1, Kling 4.0 and Vidu Q4 generate dialogue, sound effects or music with the video.

What is the best open-source video model? MiniMax H3 ranks highest, but its license excludes the US, EU, UK and South Korea. LTX-2.5 is the highest-rated open model whose license covers those regions (free for companies under $10 million in yearly revenue), and Wan 2.2 (Apache 2.0) remains popular.

What happened to Sora? OpenAI discontinued it: the Sora app closed on April 26, 2026 and the Sora API on September 24, 2026.

Is Veo 3 still available? Veo 3.1 still works in the Gemini app, Google Flow and Google Cloud. In the Gemini API, Google replaced it with Gemini Omni in October 2026.

What are Gemini Omni and HappyHorse? Gemini Omni is Google DeepMind's new video model family, launched in May 2026. HappyHorse is a closed video model from Alibaba, separate from its Wan line, that topped a leaderboard anonymously in April 2026.

How long can AI videos be? Seedance 2.5, Wan 3.0 and Kling 4.0 make up to 30 seconds in one generation. Gemini Omni can extend a scene to 40 seconds, and most other models make 5 to 15 seconds per clip.

How much does AI video generation cost? Through APIs, from about $0.01 per second at 480p to $0.60 for Veo 3.1 in 4K with audio. A 10-second 720p clip costs about $1 with Wan 3.0 or Gemini Omni and $4 with Veo 3.1 with audio. App plans start around $10 to $15 a month.

How this list was built

• The list: the companies tagged as video-generation AI models in LeadWave's database, plus the makers that lead the video leaderboards. Companies tagged as video but doing something else (video search, avatars, voice agents, interactive world models) are left out, as are platforms that host other companies' models.

• The ranking: average position on four leaderboards: LMArena text-to-video and image-to-video (dated September 21, 2026), and Artificial Analysis text-to-video and image-to-video with audio (viewed November 2, 2026). A model must be on at least three of the four; where a board hasn't rated its newest version yet, the previous version counts. Models built on another company's model are counted with the original, and models that are shut down or not yet released aren't ranked.

• The facts: release dates, clip lengths, prices and licenses checked on November 2, 2026 against each company's own pages and pricing pages, or a major news outlet.

Selling to AI companies?

LeadWave's list of AI companies covers 9,000 AI companies with CEO details. See also the best AI image generators and our list of AI labs.

About this list

Compiled by LeadWave from the public video leaderboards and company announcements, checked November 2, 2026.