Moonshot AI is a Chinese AI research company whose flagship product is Kimi, a general-purpose AI assistant. The company frames its mission around “seeking the optimal conversion from energy to intelligence” and, on its GitHub organization, states that it is “committed to solving ambitious ‘moonshot’ problems that will lead humanity to AGI” while embracing open source. Kimi is presented as a natively multimodal assistant capable of code analysis, document and spreadsheet work, and presentation creation, built on the company’s latest K-series models.
Moonshot first drew international attention as a long-context pioneer. Its early Kimi assistant was marketed on its ability to read and reason over very large inputs, such as long documents and books, at a time when most assistants had far smaller context windows. That focus shows up in the company’s open research as well: its GitHub organization publishes work like MoBA (Mixture of Block Attention) explicitly aimed at long-context language models, and Kimi-Linear, a hybrid attention architecture for handling long sequences efficiently. Long-context handling is a recurring theme across its model lineup.
The lab also releases capable open-weight models. Its Kimi-K2 model is described as an open-source Mixture-of-Experts model with one trillion total parameters and 32 billion activated parameters, reporting state-of-the-art results in knowledge, math, and coding, alongside open vision-language (Kimi-VL), audio (Kimi-Audio), and coding (Kimi-Dev) models. This puts Moonshot in the same open-release camp as other Chinese labs such as Zhipu AI, the Alibaba Qwen team, and DeepSeek.
Why business readers should care: Moonshot helped popularize the idea that an AI assistant could ingest very large documents in one pass, and its open Kimi models give organizations a long-context, agentic alternative they can self-host. It is one of the China-based labs whose open releases have intensified competition and pricing pressure across the industry.