AI Video’s New Leaders, Costs, Open Weights, and Creative Workflows

AI video has changed a lot in 12 months. The strongest models now compete on more than visual quality, with native audio becoming table stakes and typical clips running 15–30 seconds.

A ranking dated November 13, 2026, puts an open-weights model in the top three. It also places quality, API price, and the license status of model weights in the same conversation, giving creators and developers more than one way to judge a video system.

How the 2026 leaderboard measures quality

The ranking uses one quality signal: blind human-preference Elo from the Artificial Analysis Video Arena. Artificial Analysis ranks models through blind pairwise votes from validated evaluators, with each vote comparing 2 clips created from the same prompt.

The main board is AA-Video-T2V v2.1 with audio, and the ranking also reports the AA-Video-I2V v2.1 score. Kling 3.0 1080p (Pro), developed by Kuaishou, serves as the anchor at 1000 Elo on both boards. A score below 1000 means voters preferred Kling 3.0 more often.

The evaluation included 7,651 votes. Most top-10 scores carry a ±13 to ±15 Elo interval, so small gaps do not tell the whole story. Models within about 25 points are statistically close, which means ranks 3–8 work better as a cluster than as a strict order.

The models leading the field

Wan 3.0 from Alibaba ranks first, followed by Utopai X from Utopai Studios. MiniMax H3 (768p) from MiniMax takes third place, putting an open-weights model in the top three without treating that status as a separate quality category.

Dreamina Seedance 2.5 from ByteDance Seed ranks fourth, while Gemini Omni Flash 1.2 from Google ranks fifth. FLUX 3 Max, listed with Max and fal × Black Forest Labs, holds sixth place, followed by HeyGen Video 1 from HeyGen.

MiniMax H3 Max fal (H3 post-train) ranks eighth, SkyReels V4 from Skywork AI takes ninth, and FLUX 3 Black Forest Labs appears at number 10. The wider table also lists Seedance 2.0, P-Video-2-Pro from Pruna AI × MiniMax, Pika H3 Flex from Pika (H3 post-train), Veo 3.1 and Veo 3.1 Lite from Google, and LTX-2.5 Fast from Lightricks.

That spread matters because the list is no longer a simple contest between closed systems. It places model quality beside access, price, and weight licensing, giving teams a clearer way to think about deployment choices. Deployment options and specs can also be checked through the GPULeaderboard.com chrome extension.

Price, hardware, and the practical tradeoffs

Artificial Analysis lists cost per minute on each creator’s API, then normalizes to 1080p where that option exists. The ranking divides that figure by 60 to calculate price per second. Vendor rates at 720p are often lower, and the comparison flags those gaps instead of treating every price as a direct match.

The figures in the comparison include 93 credits per second and $15 for 1,000 credits. Other listed specifications include a 33B model, 32 kHz stereo audio, and 2 task-specific checkpoints totaling 288 GB. Those details matter when a team weighs API access against running model weights through its own deployment setup.

Native audio changes the basic question for creators. A video model is no longer judged only on moving images when clips can include sound and run for 15–30 seconds. Prompt design also matters, since the same system can respond to different instructions about a scene, camera, motion, lighting, and sound.

Tools moving from clips to complete workflows

PixVerse has launched PixVerse text to video, which turns written prompts into moving footage. Its workflow asks users to write a scene and add direction for the camera, motion, lighting, and sound, without a film crew, an editing timeline, or a complicated production setup.

PixVerse treats prompt writing as a craft of its own. Its recommended structure is a short directing note built from subject, setting, action, camera, light, sound, and format. Subject and setting can include people, products, rooms, landscapes, props, and avatars.

YouArt is taking a different route with YouArt Desktop App, an AI video agent for Mac and PC. YouArt, an AI-powered creative platform backed by Y Combinator, says the app works with local project files, professional creative applications, and team feedback.

Unlike video generators focused on individual clips, YouArt Desktop App can understand existing projects, operate editing software, and handle production tasks from initial research to final export. When a client sends feedback through Slack or Gmail, the app can bring it into the project canvas as a comment pin.

The leaderboard shows where video quality stands on November 13, 2026. PixVerse and YouArt show how the surrounding workflow is changing too: from writing a directing note, to generating footage, to managing feedback and finishing a project inside creative software.

Stay connected via Google News

Follow us for the latest AI News