MiniMax-H3 Claims #1 Open Model Spot on Video Arena Leaderboard
MiniMax-H3 has become the top-ranked open-weights model on the Video Arena leaderboard for both Text-to-Video and Image-to-Video tasks. It outperformed the next best open model, Hunyuan-Video-1.5, by over 280 points in both categories. This milestone demonstrates a significant leap in open-source video generation capabilities, closing the performance gap with proprietary models. It provides developers and researchers with a highly competitive, accessible tool for high-quality video synthesis. MiniMax-H3 is an omni-modal model capable of generating up to 15-second 2K resolution videos with native stereo audio from unified text, image, video, and audio contexts. In the Image-to-Video Arena, it scored 1,455 points, placing it just 3 points behind the third-place overall model, Muse Video.
## BACKGROUND
The Video Arena, hosted by Arena.ai, is a crowdsourced benchmarking platform where users compare side-by-side video generations from different AI models to rank them. Open-weights models allow the public to download and run the model weights locally, contrasting with closed-source, API-only models.