Recently, SuperCLUE released its latest updated overall leaderboard for I2V image-to-video models. In this evaluation, which has drawn wide attention from the industry, MiniMax H3, launched by Xiyu Technology, took first place on the domestic leaderboard with a total score of 71.46, and also achieved a high score of 75.82 in the application capability dimension, demonstrating extremely strong real-world performance.
According to the rules of this leaderboard, overseas participating models serve only as a reference control group and are not counted in the official rankings, while models with a score gap within 1 point are judged to be tied. In the specific rankings, Alibaba Wan3.0 ranks second domestically with a total score of 66.63, with a basic capability score of 68.51 and outstanding performance on individual metrics; closely followed, ByteDance Seedance2.5 and Kuaishou Kling V3Pro tied for third domestically with scores of 64.63 and 64.06 respectively; Aishi Technology PixVerse V6 and Shengshu Technology Vidu Q3Pro jointly rank in the fourth tier. In addition, among the overseas models in the reference group, Google Gemini Omni1.1Flash scored a total of 67.38, and X.AI's Grok Imagine Video1.5 scored 65.79.
It is worth noting that some participating models were subject to certain rule restrictions during testing. For example, models such as Seedance2.5, Gemini Omni1.1Flash, and PixVerse V6 triggered constraints such as real-person appearance limitations, generation duration caps, or content moderation, which caused some test questions to fail to produce results and be counted as 0 points, objectively affecting their final total scores to a certain extent.
Judging from the results of this SuperCLUE evaluation, the overall technical level of current domestic AI video models has successfully caught up with overseas peers, and has surpassed them on some individual metrics in specific application areas. However, industry experts also point out that benchmark test results can only serve as an objective technical baseline reference, and the specific video generation quality and practicality still need to be comprehensively assessed by creators based on their own usage scenarios and actual needs.