MiniMax leads with an 84.5% implied probability due to the strong trader consensus around its H3 omni-modal model, released in late July and open-sourced in early August 2026. The 33B-parameter system excels at image-to-video by unifying text, up to nine reference images, video clips, and audio inputs to generate up to 15-second clips at 768p or 2K with native stereo sound and lip-sync, outperforming rivals on recent leaderboards for motion coherence and multimodal control. Optimizations like H3 Max have accelerated inference while preserving quality, driving ecosystem adoption in tools such as ComfyUI. Google’s Veo 3.1 holds secondary positioning at 12.5% for its cinematic polish, but shorter default lengths limit its edge; other labs trail as the October 31 resolution nears with no major counter-releases reported.