O conteúdo está em inglês; a navegação e os filtros mantêm o idioma selecionado.

MiniMax H3 launches with open weights, but benchmark leadership varies by task

MiniMax has released H3 with downloadable weights. Arena.ai places it at the top among open models in its text-to-video and image-to-video arenas, while Artificial Analysis reports different rankings by task.

Publicado em 06/08/2026, 09:18:03Última atualização 06/08/2026, 09:18:03Publicação editorial

MiniMax has released its H3 video model with downloadable weights. The multimodal system accepts text, images, video, and audio, and can generate short clips with native audio. Arena.ai reports that H3 leads the open-model field in both its Text-to-Video and Image-to-Video arenas, citing scores of 1,455 and 1,476 respectively.

The results are not consistent across evaluation platforms. Artificial Analysis ranks H3 first in Video Editing, second in Text-to-Video, and third in Image-to-Video. The Decoder has reported a similar breakdown. As a result, describing H3 as the unqualified overall SOTA model across both benchmark systems would be too broad; the available evidence supports a more limited claim that it is a leading open-weight video model and tops some task-specific rankings.

The open release gives developers more options for local deployment and workflow integration. Its standing should still be interpreted in light of each leaderboard’s methodology, task mix, sample set, and evaluation date.

Evidências das fontes 6

A verificação de fontes não garante exatidão permanente. Considere a fonte original, a data e as correções posteriores.

MiniMax launches open-weight H3 and claims video benchmark lead - RuntimeWireruntimewire.com · Complementar

MiniMax's H3 AI model performing at the top of a benchmark evaluation (Isometric 3D render with paper-cut materials, featuring chunky low-poly shapes and matte paper textures.) MiniMax, founded by Yan Junjie, said days after launching H3 that the model had become the highest-ranked open-weight video model on both Arena and Artificial Analysis. MiniMax on X [...] MiniMax's H3 blog frames the release as part of a push to "break task boundaries" across text, image, video and audio. The release's practical distinction is downloadable weights: the Comfy-Org Hugging Face model card provides weight-download instructions, and ComfyUI's H3 documentation gives developers local workflow examples. ### The public evidence is narrower than MiniMax's claim MiniMax's benchmark post says H3 is the leading open video-generation model on Artificial Analysis. MiniMax made the same claim for Arena. The Artificial Analysis leaderboard reviewed for this story listed H3 at No. 2 overall behind Google Gem