Moonshot AI logo

Moonshot AI

Chinese AI startup behind the Kimi chatbot and Kimi K2.5 (1T MoE, 32B active, 76.8% SWE-bench Verified — best open-source coding model). Agent Swarm orchestrates up to 100 sub-agents. Kimi Code IDE. Modified MIT license.

Share

Listen to this lesson

Free preview · first 0:30
0:00 / 0:30

Unlock audio and more

Audio streaming, downloadable PDFs and certificates come with Plus and Pro.

📋About Moonshot AI

Updated September 12, 2026

Moonshot AI (月之暗面) is a Beijing-based frontier AI lab founded in March 2023 by Yang Zhilin, previously a co-founder of Zhipu AI. The company became China's fastest decacorn, reaching a $10 billion valuation in February 2026, and has raised through a rapid succession of rounds since — a $500 million Series C in January 2026 from IDG Capital, Alibaba, and Tencent, then a $2 billion round in May 2026 led by Meituan's Long-Z Investments arm with Tsinghua Capital, China Mobile, and CPE Yuanfeng participating. As of July 2026 the lab is reported to be raising again at a materially higher valuation.

Moonshot's product line runs on the Kimi brand: the Kimi chatbot on web and mobile, Kimi Code (a coding agent with terminal, VS Code, Cursor, and Zed integrations, plus MCP support), and Kimi Work, a desktop agent for macOS and Windows that runs the lab's Agent Swarm system on local hardware. The chatbot first gained prominence on long-context document analysis, and long context has remained a defining thread through every generation since.

The company's strategic identity is open weights at frontier scale. Where most labs treat open release as a tier below their flagship, Moonshot has released its frontier models publicly — the K2 family shipped under a Modified MIT License — and priced API access aggressively enough to exert real cost pressure on US frontier vendors. The lab is one of the few Chinese frontier labs to publicly disclose consumption figures, reporting $200 million in annualized recurring revenue in April 2026, and Kimi models have consistently ranked among the most-used on OpenRouter.

On July 16, 2026, Moonshot shipped Kimi K3, the clearest expression of that strategy to date. K3 is a 2.8 trillion parameter mixture-of-experts model activating just 16 of 896 experts per token, built on the lab's own Kimi Delta Attention and Attention Residuals work, with a 1 million token context window and native vision. It is the largest open-weights model any Chinese lab has produced, and it beats Claude Opus 4.8 on most of the published coding and agentic suite — most dramatically on FrontierSWE, 81.2 against 66.7. Moonshot concedes K3 still trails Claude Fable 5 and GPT-5.6 Sol overall, and names user experience rather than raw capability as the remaining gap. Moonshot published the full weights on July 27, 2026 — 96 safetensors shards spanning well over a terabyte — but under a custom Kimi K3 License rather than the Modified MIT terms the K2 family used. The new license permits download, modification, and commercial use with conditions: an inference or fine-tuning service whose revenue exceeds $20 million over 12 consecutive months must negotiate a separate agreement with Moonshot, and any product above 100 million monthly active users or $20 million in monthly revenue must display Kimi K3 branding in its interface. Internal use and access through Moonshot’s own products are exempt. The shift matters beyond the legal text: the lab whose strategic identity is open weights at frontier scale attached commercial strings at exactly the point its model became competitive with the proprietary frontier.

K3's pricing marks a shift in posture. At $15 per million output tokens, Moonshot is no longer undercutting the frontier the way earlier Chinese open models did — it is pricing like a frontier lab, with the aggressive rate reserved for cached input at 30 cents per million tokens. Taken with the launch's timing one day before Xi Jinping used the World AI Conference in Shanghai to make open source explicit Chinese national strategy, Moonshot has become the most concrete evidence available for a policy position China's head of state is now articulating directly.

On September 10, 2026 Anthropic alleged in a threat intelligence report that Moonshot served Claude to its own customers without telling them. According to that account, Moonshot forwarded customer requests to Claude instead of processing them with Kimi, displayed Claude's answers as Kimi's, and kept at least some of those exchanges to train its own models — in one ten-day period roughly 300,000 requests, the majority routed to Opus, through a network of 5,380 fraudulent accounts mostly appearing to sit in Singapore and Japan, with more than 23 million exchanges attributed between May and July 2026. Anthropic says Moonshot defeated a specific control by saving the reasoning signature Claude returns in place of raw thinking, opening a new session, and eliciting the model to convert it back into the full trace. It adds that the relayed sessions contained sensitive material from Moonshot's users, including one it assesses was likely affiliated with the People's Liberation Army analyzing surveillance footage. Moonshot had not publicly responded at publication, and Anthropic competes with it directly.

🛠️Products & Tools (5)

Kimi CodeAI Coding

Moonshot AI's open-source coding tool with integrations for terminals, VS Code, Cursor, and Zed. Companion to the Kimi K2.5 model.

Kimi K2.6Foundation Models & Open Source

Moonshot AI's flagship before Kimi K3 — a 1 trillion parameter MoE model (32B active) with 262K context, native INT4 quantization.

Kimi K3Foundation Models & Open Source

Moonshot AI's flagship — a 2.8 trillion parameter mixture-of-experts model activating 16 of 896 experts per token, with a 1 million token context window and native vision. Strong on coding benchmarks; the largest open-weights model from a Chinese lab, with weights published in July 2026 under a custom Kimi K3 License.

Kimi K2.5Foundation Models & Open Source

Moonshot AI's 1 trillion parameter MoE model (32B active). 256K context, multimodal (15T mixed tokens). Beats GPT 5.2 on SWE-Bench Multilingual and Claude Opus 4.5 on VideoMMU.

KimiInternational Chatbots

Moonshot AI's assistant, powered by the Kimi K3 family — a 2.8 trillion parameter mixture-of-experts model with a 1 million token context window and native vision. Kimi Code adds IDE integration. Named by Anthropic in September 2026 over alleged unauthorized distillation of Claude.

Keep track of the companies you’re watching

  • The AI Hub on a phone: a 12-day AI Skill Streak and an expanded Content updates alert listing the saved items that changed.
  • Recommended for you on a phone: nine personalised suggestions labelled Trending in AI news, On your saved list, and Popular.
  • My AI Tools on a phone: saved tools including GitHub Copilot and OpenAI Codex, each with an Updated badge.

Swipe for Recommended for you and My AI Tools

Your AI Hub — sample data.

📰Moonshot AI in the News

View all 26 Moonshot AI stories in Top AI Stories

Learn more about Moonshot AI

Our curriculum includes an in-depth overview of Moonshot AI's strategy, models, and competitive positioning.

View Moonshot AI Overview Lesson