---
id: 20260404-T0-01
title: "Dynin-Omni：首个多模态扩散语言模型"
title_en: "Dynin-Omni: First Omnimodal Diffusion Language Model"
url: https://ai.daily.yangsir.net/daily/20260404-T0-01
issue_date: 2026-04-04
publish_date: 2026-04-03T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2604.00007
---

# Dynin-Omni：首个多模态扩散语言模型

研究人员发布Dynin-Omni，首个基于掩码扩散的多模态统一模型。该模型能同时处理文本、图像、语音理解和视频分析，性能超越现有自回归模型。在多模态基准测试中，准确率提升22%，推理效率提高30%。

## English Version

**Dynin-Omni: First Omnimodal Diffusion Language Model**

Researchers released Dynin-Omni, the first masked-diffusion-based omnimodal model unifying text, image, speech, and video understanding. It outperforms autoregressive models with 22% higher accuracy and 30% improved inference efficiency in multimodal benchmarks.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2604.00007)

**详情页**：https://ai.daily.yangsir.net/daily/20260404-T0-01

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*