AI 导读
商汤开源了多模态模型 SenseNova U1，其推文称这是首个在单一架构中实现的真正统一多模态模型，不含 VAE、视觉编码器和适配器，进行纯像素与词元推理。该推文还称 8B 版本在速度和质量上均超过 20B+ 竞品。
正文
🚨 BREAKING: SenseTime open-sourced SenseNova U1, the first truly unified multimodal model in one architecture.
No VAE, no visual encoder, no adapter mess. Pure pixel-word reasoning.
The 8B variant is now beating 20B+ competitors on both speed and quality :
来源：@Ubermenscchh · x.com