The ranking of artificial intelligence video generation models has changed significantly in a year. As of November 13, 2026, Alibaba’s Wan 3.0 holds the top spot in the text-to-video ranking, and an open weights model (whose parameters can be downloaded) is in the top 3.
This ranking is based on a single quality criterion, an Elo score (a rating system based on pairwise comparisons) calculated from blind human votes, drawn from the Artificial Analysis Video Arena. It is cross-referenced with the API price per second and the license status of the weights. According to the source, native audio is now an essential prerequisite, and clips last 15 to 30 seconds.
The leading models
Wan 3.0 scores 1175 points in text-to-video, with 7,651 votes. Launched on August 24, 2026, it generates 1080p, 30-second clips from text, PDFs, or presentations, priced at $0.20 per second. Its weights are not downloadable, and Wan 2.2, released in July 2025, remains Alibaba’s last major open video model. In image-to-video, it drops to 11th place, with 1108 points.
Utopai X, with 1170 points, is the only model appearing in the top 2 of both rankings. It was designed by Utopai Studios, a film studio based in Mountain View, which obtained it by post-training MiniMax H3. It is only accessible via the PAI platform, billed at 93 credits per second, with packages starting at $15 for 1,000 credits, and has no API for developers.
MiniMax H3 comes in third, with 1161 points and a price of $0.08 per second. Its weights are published under the MiniMax H3 Community License, but this license blocks local deployment in the United States, the European Union, the United Kingdom, and South Korea. Only the H3-Base version is open, with a cap of 768p, and the repository weighs 288 GB. Several models in the ranking, including Utopai X and HeyGen Video 1, are post-trained versions of H3.
In image-to-video, the top spot goes to ByteDance’s Dreamina Seedance 2.5, with 1161 points. Artificial Analysis shows $34.12 per minute, the highest price in the ranking, while BytePlus advertises around $0.23 per second for a specific format, which advises checking the access conditions in your region.
Price and margin of error
Quality and price no longer go hand in hand. HeyGen Video 1 (1152 points, $0.04 per second) and Pika H3 Flex (1124 points, $0.035 per second, released October 29, 2026) are among the cheapest in the top 10. Google’s Gemini Omni Flash 1.2 scores 1154 points for $0.104 per second. According to the source, Wan 3.0 yields about 23 more points than HeyGen Video 1 for a cost 5 times higher per second.
The scores have a margin of error of ±13 to ±15 Elo. Models within 25 points of each other are statistically close, so ranks 3 to 8 form a group rather than a strict order. Google’s Veo 3.1 scores 964 points, below the anchor set at 1000 for Kling 3.0 1080p Pro. Kling 4.0, available on klingai.com with native 30-second clips, does not yet appear in Artificial Analysis rankings.
