H3 Image Matches GPT on Style, but Not Complex Prompts
TMTPost tested MiniMax’s open-weight H3 Image against GPT Image 2.6 across six image-generation tasks. H3 Image delivered stronger realism and aesthetics in several tests, but GPT Image 2.6 was more reliable for detailed instructions, precise quantities, commercial layouts and multi-reference composition. MiniMax claims H3 Image can generate a roughly 16-megapixel 4K image in as little as 10 seconds for less than RMB 0.30, while the reported test also found its pricing and generation speed competitive.
The comparison indicates that H3 Image is a credible open-weight option for high-volume visual creation, while GPT Image 2.6 retains an advantage where exact execution and production-ready design matter.
(本文作者为 第四波,钛媒体经授权发布)
文 | 第四波,作者|冯皓钦
前两天,闫俊杰刚刚说,自己最近全部精力都投注在M3 Pro上。结果11日凌晨,MiniMax就先扔出一道前菜——H3 Image,并同步开放了模型权重,坚定地践行着自己的多模态路线。
在正式评测之前,我们先了解一下H3 Image几个核心功能点。
它衍生自此前的H3多模态模型,除了文生图、图片编辑,还支持最多16张参考图输入。官方重点强调的几个优势:生成速度、画面真实感,复杂指令遵循能力以及便宜的价格。一张约1600万像素的4K图片,最快只需要10秒,价格不到3毛钱。
但在前几天,11月4日,OpenAI端出了GPT Image 2.6,也重点升级了生成速度、图像编辑和视觉一致性。两款模型前后脚发布,难免让人好奇:MiniMax这次到底追上GPT了吗?