
Independent company · Beijing, China
Moonshot AIKimi
Beijing lab behind the Kimi models, whose open-weight Kimi K3 family at 2.8 trillion parameters is the largest open release to date.
Beijing, China
- Founded
- 2023
- Type
- Independent company
- Based in
- Beijing, China
- Key papers
- 3
- Featured projects
- 8
About
Moonshot AI was founded in Beijing in March 2023 by Yang Zhilin, Zhou Xinyu and Wu Yuxin, who had been schoolfriends at Tsinghua University. Yang previously worked at Google Brain and Meta. The company is named after Pink Floyd's The Dark Side of the Moon.
It operates the Kimi assistant, an API platform and the Kimi Code command-line agent, and publishes model weights on Hugging Face. Kimi K2 appeared in July 2025, K2.5 added native vision in January 2026, K3 followed in mid-2026 and Kimi K3.1 Pro was published in November 2026.
Moonshot is venture-backed, with Alibaba and Tencent among its investors, and releases K3 under a custom license that requires revenue sharing from very large commercial inference providers.
Mission and ambitionsMoonshot states its aim as building foundation models toward artificial general intelligence, sharing research with the open-source community.
Known for
- 01
Kimi open-weight model family
- 02
Very large mixture-of-experts open models
- 03
Long-context chat assistant
- 04
Agent Swarm and Kimi Code
Milestones
- 2023
Kimi chatbot
Moonshot released the first version of Kimi in October 2023, capable of long Chinese-language context.
- 2025
Kimi K2
The K2 weights were published on Hugging Face in July 2025.
- 2026
Kimi K2.5
A multimodal upgrade with a 400-million-parameter MoonViT vision encoder and Agent Swarm, released in January 2026.
- 2026
Kimi K3
A 2.8-trillion-parameter open-weight model family launched in July 2026.
- 2026
Kimi K3.1 Pro
A model with a one-million-token context window published in October and November 2026.
Open source
Archived legacy Kimi CLI, the most starred repo.
Repository for the Kimi K2 model series.
Repository for Kimi K3.
Kimi Code CLI for agentic coding.
- Kimi-K3model
Most downloaded model in the org.
- Kimi-K2.5model
Kimi K2.5 release.
- Kimi-K2-Instructmodel
Instruction-tuned K2.
- PerceptionBenchdataset
Multimodal perception benchmark.
Research papers
Notable releases
Openness
Open weights on Hugging Face; exact licence stated on each model card.
People
- Yang ZhilinCo-founder and CEO
- Zhou XinyuCo-founder
- Wu YuxinCo-founder
Listed from the lab's own pages and public profiles.
Contact and community
Reach the team through its website, GitHub issues or Hugging Face discussions.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



