Meituan open-sources omni-modal model LongCat-Flash-Omni
Original title:meituan-longcat/LongCat-Flash-Omni-FP8
AI summary
Meituan has open-sourced the omni-modal model LongCat-Flash-Omni, using a 560-billion-parameter MoE architecture (27 billion active) and supporting 128K context and real-time audio-video interaction. Based on shortcut-connected MoE and zero-computation experts, it is equipped with lightweight codecs and a chunked feature interleaving mechanism, improving efficiency through curriculum-style progressive training. It surpasses Qwen3-Omni and Gemini-2.5-Pro on benchmarks such as OmniBench and WorldSense, and reaches leading levels in tasks including document understanding, speech recognition, and GUI control.
Key Takeaways
- Meituan has open-sourced the omni-modal model LongCat-Flash-Omni,using a 560-billion-parameter MoE…
- Based on shortcut-connected MoE and zero-computation experts,it is equipped with lightweight…
- It surpasses Qwen3-Omni and Gemini-2.
- 5-Pro on benchmarks such as OmniBench and WorldSense,and reaches leading levels in tasks including…
Quick Facts
- Published
- Updated
- Source
- 美团 LongCat:HuggingFace 新模型
- Category
- Other
美团开源 560B 参数多模态模型,27B 激活即可实现实时音视频交互
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Meituan open-sources omni-modal model LongCat-Flash-Omni. October 24, 2025. https://new.nat6.net/en/news/cmocymrfh005mslqln484ev5b/