Moonshot AI launched an open-source AI model, Kimi K2, built to handle frontier knowledge, maths, coding, and agentic tasks. The Beijing-based startup is pushing the edge against competitors like DeepSeek.
Kimi K2 uses a mixture-of-experts (MoE) architecture. It packs 1 trillion parameters in total, with 32 billion activated parameters – specialized units that engage for specific tasks. This approach cuts pre-training computation costs and speeds up performance during use, Moonshot said Friday.
Moonshot released two open versions: Kimi-K2-Base for researchers wanting full control and fine-tuning, and Kimi-K2-Instruct for ready-to-go chat and agentic AI. 
The model is available now through Moonshot’s web and mobile apps.
Moonshot’s move adds to a trend of open sourcing in AI, aimed at boosting efficiency and adoption. Other players like Baidu, Alibaba Cloud, and DeepSeek are also riding this wave.
Moonshot explained the benefits of open source:
“The open-source approach gives public access to a program’s source code, allowing third-party software developers to modify or share its design, fix broken links or scale up its capabilities.”
