---
type: "Topics"
locale: "en"
url: "https://longbridge.com/en/topics/42966132.md"
description: "📢 𝗝𝗨𝗦𝗧 𝗜𝗡: Kimi AI Unveils Kimi K3 With 2.8 Trillion-Parameter MoE Architecture👉 𝗞𝗲𝘆 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀:➤  𝗞𝗶𝗺𝗶 𝗔𝗜 launched its flagship 𝗞𝗶𝗺𝗶 𝗞𝟯 foundation model.➤  Kimi says its new architecture delivers 𝟮.𝟱𝘅 more intelligence per unit of compute.➤  𝗞𝗶𝗺𝗶 𝗞𝟯 features a 𝟮.𝟴 𝘁𝗿𝗶𝗹𝗹𝗶𝗼𝗻-𝗽𝗮𝗿𝗮𝗺𝗲𝘁𝗲𝗿 Mixture-of-Experts architecture.➤  Model supports a 𝟭,𝟬𝟬𝟬,𝟬𝟬𝟬-𝘁𝗼𝗸𝗲𝗻 context window.➤  Kimi K3 includes 𝗻𝗮𝘁𝗶𝘃𝗲 𝘃𝗶𝘀𝘂𝗮𝗹 understanding capabilities.➤  Company released 𝗺𝗼𝗱𝗲𝗹 𝘄𝗲𝗶𝗴𝗵𝘁𝘀 and a technical report.➤  Kimi will open-source high-performance attention compute kernels.➤  Open components also include an 𝗠𝗼𝗘 communication library and agent infrastructure.➤  Release targets large-scale 𝗔𝗜 𝗮𝗴𝗲𝗻𝘁 development and deployment.👉 𝗪𝗵𝘆 𝗧𝗵𝗶𝘀 𝗠𝗮𝘁𝘁𝗲𝗿𝘀:➤  Signals intensifying competition in advanced foundation models.➤  Focuses on improving AI efficiency instead of only increasing model size.➤  Open-sourcing key infrastructure could accelerate AI ecosystem innovation.➤  Large context and multimodal capabilities strengthen enterprise AI use cases."
datetime: "2026-07-27T15:24:23.000Z"
locales:
  - [en](https://longbridge.com/en/topics/42966132.md)
  - [zh-CN](https://longbridge.com/zh-CN/topics/42966132.md)
  - [zh-HK](https://longbridge.com/zh-HK/topics/42966132.md)
author: "[Hardik Shah](https://longbridge.com/en/profiles/27423834.md)"
---

# 📢 𝗝𝗨𝗦𝗧 𝗜𝗡: Kimi AI Unveils Kimi K3 With 2.8…