Moonshot AI is a Beijing lab founded in 2023 by Yang Zhilin. It ships Kimi K3, a 2.8T-parameter open-weight model with a 1M-token context window, plus K3.1 Pro as of October 24, 2026. Valued at $20B in May 2026; a later $35B figure is only reported.
Moonshot AI is a Beijing AI lab, founded in March 2023 by Yang Zhilin, that makes the Kimi assistant and the open-weight Kimi K3 family. K3 has 2.8 trillion parameters and a 1-million-token context window. Its May 2026 Series D valued it at about $20 billion, led by Meituan's Long-Z Investments.
Founded: 2023 · HQ: Beijing, China · Team: 200-300 · CEO: Yang Zhilin · Funding: $2B Series D in May 2026 with Long-Z Investments (Meituan) as lead; a later round above $3.5B was reported in July 2026 and is unconfirmed · Valuation: $20B at the May 2026 Series D; $35B reported by Yicai (July 30, 2026, unnamed sources, unconfirmed)
About Moonshot AI
Moonshot AI was founded in March 2023 in Beijing by Yang Zhilin together with Tsinghua University schoolmates Zhou Xinyu and Wu Yuxin. Yang had worked as an NLP researcher across Carnegie Mellon and Google Brain, and as first author of XLNet he helped build a pretraining model that beat Google's BERT on 20 of 20 benchmark tasks. The company is known in Chinese as Yuezhi Anmian (Dark Side of the Moon). It set out to build long-context, agentic language models at a time when most Chinese labs were shipping short-context chat assistants.
The product line now runs on Kimi K3. Moonshot released Kimi K3 on July 16, 2026 as a 2.8-trillion-parameter model with 104 billion activated parameters, native vision and a 1-million-token context window, and published the weights on Hugging Face on July 27 under its own Kimi K3 License. On October 24, 2026 it followed with Kimi K3.1 Pro, a post-trained version of K3 with a Highspeed variant that Moonshot says runs 2 to 3 times faster; the weights are promised by November 4. For K3.1 Pro the company reports 65.7% on Terminal-Bench 4.0, against 66.4% for Claude Opus 5.5 and 58.2% for GPT-6 Astra, which are Moonshot's own figures and not independently verified. The earlier Kimi K2.7 Code and its Highspeed variant remain listed, but the Kimi API platform says K2 series models close to newly registered users on October 31, 2026 and are planned for retirement by the end of December 2026.
Consumer access is through Kimi, the chat assistant, plus Kimi Work and the Kimi Code coding agent. Developers pay per token on the Kimi API platform, with no subscription. Per-token rates run $2.00 input and $10.00 output for K3.1 Pro ($0.20 on a cache hit), $3.00 and $15.00 for K3 ($0.30 cached), and $0.95 and $4.00 for K2.7 Code ($0.19 cached). Pricing rose sharply with K3 compared to the K2 generation, which changes the cost argument Moonshot used to make against closed labs, with Anthropic the usual yardstick and OpenAI the other. Third-party hosts such as Together AI carry the open weights for teams that prefer not to call Moonshot directly.
Funding has moved quickly. The May 2026 Series D brought in $2 billion at a roughly $20 billion valuation, with Meituan's venture arm Long-Z Investments in front and Alibaba, Tencent, HongShan and IDG Capital among existing backers. Yicai wrote on July 30, 2026, citing unnamed sources, that a later round worth over $3.5 billion put the valuation at $35 billion, with another raise near $50 billion pre-money in discussion. Moonshot had not commented, so treat those two figures as reports rather than confirmed terms.
Moonshot competes most directly with other Chinese open-weight labs. DeepSeek is the closest rival on per-token price, Z.ai (see its profile) pushes hard on coding agents, MiniMax leans toward video and consumer scale, and the Qwen team sits inside Alibaba Cloud. Against that group its distinguishing bets are very large open models and long-horizon agentic coding, and its own launch post says K3 still trails the strongest proprietary models. For a model-by-model comparison against a closed rival, see ChatGPT vs Kimi.
Where the alternatives win: teams with strict compliance needs should look elsewhere or self-host. Moonshot lists no third-party security certification and no trust center that we could find. Its privacy policy says user content, including prompts and uploads, is used to optimize and train its models, and it states that data is held on servers in Singapore. It has not published a responsible scaling policy or a formal safety system card for the K3 generation. A buyer who needs a zero-retention contract should start with Google, or run the open weights inside their own environment, which at 2.8 trillion parameters is a serious hardware commitment.
Moonshot publishes openly. Technical reports and model cards appear on Hugging Face, and the research page lists Kimi K3, PerceptionBench and Kimi K2.6 as recent releases. Its Kimi Code Bench suites are internal, so the headline coding gains it quotes should be read as vendor-reported until a third party reproduces them.
Mission
Build long-context, agentic large language models and release them openly to push the boundary of what AI assistants can do for everyday knowledge work.
Products
- Kimi (Consumer chat assistant): https://www.kimi.com
- Kimi K3 (Open-weight flagship model): https://huggingface.co/moonshotai/Kimi-K3
- Kimi K3.1 Pro (Flagship model with Highspeed variant): https://www.kimi.com/en/blog/kimi-k3-1-pro
- Kimi API Platform (Developer API): https://platform.kimi.ai
- Kimi Code (CLI / coding agent): https://www.kimi.com/code
Moonshot AI Models on HokAI
Links
Frequently Asked Questions
What does Moonshot AI build, and which Kimi model is current?
Moonshot AI builds the Kimi chat assistant, the Kimi Code coding agent, the Kimi API platform and the open-weight Kimi model family. The current flagship is Kimi K3, released July 16, 2026, with 2.8 trillion total parameters, 104 billion activated per token and a 1-million-token context window. On October 24, 2026 Moonshot added Kimi K3.1 Pro and a Highspeed variant, with weights promised by November 4. K3 weights are published on Hugging Face under the Kimi K3 License, which is Moonshot's own license rather than the Modified MIT terms used for the earlier K2 models.
Who founded Moonshot AI and who leads it?
Yang Zhilin founded Moonshot AI in March 2023 in Beijing with Tsinghua schoolmates Zhou Xinyu and Wu Yuxin, and he remains CEO. Before the company he did NLP research across Carnegie Mellon and Google Brain and was first author of XLNet, which beat BERT on 20 of 20 benchmark tasks in its paper. No separate president or chief safety officer has been publicly named.
How much funding has Moonshot AI raised and what is it valued at?
The best-documented round is a $2 billion Series D on May 7, 2026, led by Long-Z Investments (Meituan's venture arm) at roughly a $20 billion valuation, with earlier investors such as Alibaba and Tencent also in the cap table. A Yicai report dated July 30, 2026 relied on unnamed sources to say a later round above $3.5 billion lifted the valuation to $35 billion and that a raise at about $50 billion pre-money was being pursued. Moonshot had not commented on those two figures, so they are unconfirmed.
How much does the Kimi API cost per million tokens?
The Kimi API platform is pay-as-you-go with no subscription. Per million tokens, Kimi K3.1 Pro lists at $2.00 input, $10.00 output and $0.20 cache hit, and $3.00, $15.00 and $0.30 for Kimi K3. Kimi K2.7 Code lists at $0.95 input, $4.00 output and $0.19 cache hit. Highspeed K3.1 Pro doubles the input and output rates to $4.00 and $20.00. The platform says K2 series models close to new users on October 31, 2026 and retire by the end of December 2026.
Is Moonshot AI safe to use for sensitive or regulated data?
Moonshot publishes no third-party security certification and we found no trust center. Its privacy policy says user content such as prompts and files is used to optimize and develop its models, and that its data is held in Singapore. No responsible scaling policy or formal safety system card has been published for K3. For regulated work, self-hosting the open weights or choosing a vendor with audited controls is the safer route.
Top Alternatives
HokAI guides that mention Moonshot AI
- Z.ai (GLM) Alternatives in 2026: Coding Plans, APIs and Older GLM Models: Z.ai and GLM alternatives in 2026: Kimi Code and MiniMax's M Plan against the GLM Coding Plan, DeepSeek and MiniMax for the API, and what replaces old GLM.
- MiniMax Alternatives in 2026: The Best Pick for Each MiniMax Product: MiniMax alternatives in 2026, by product: DeepSeek and GLM for the M3 API, the GLM Coding Plan for coding, ElevenLabs for voice, and why video should stay.
- Best AI for Coding Unity Games in 2026: Which AI is best for coding Unity games? We compare Unity's own assistant, its free CLI and plugin, Cursor, Copilot and Rider on cost, fit and risk today.
- Kimi vs ChatGPT: Is Kimi K3 Better, and for Whom?: Kimi or ChatGPT? Kimi K3 is open-weights, with plans from $19 a month. ChatGPT Plus is $20 with GPT-6.1 Sol, which scores higher and costs less per token.
- Harvey vs Legora: The 35-Hour Onboarding vs the 6-Month Build: Harvey raised at $15.5B in Sept 2026; Legora's last mark is $5.6B. Neither lists prices, so we compare seat cost, onboarding time and data handling.
- What Mistral Large 4 Is Actually For: Mistral Large 4 is a 1.05T open-weight preview with a halved sale price. Who it suits, what it costs, and what is still unknown about weights and licence.