What Is Moonshot AI? Company, Kimi AI & Everything You Need to Know

What Is Moonshot AI? The Company, Kimi Assistant & Everything You Need to Know

Moonshot AI is a Beijing-based artificial intelligence company, founded in March 2023, that builds the Kimi AI assistant and a family of open-weight large language models. The company has raised a total of $3.77 billion over 4 funding rounds, and its valuation stands at $20 billion. Its flagship product, Kimi, was the first AI chatbot to support 128,000-context processing and has since scaled to handle millions of characters per conversation, with its latest model, Kimi K3, reaching 2.8 trillion parameters. Keep in mind that Moonshot AI operates primarily in the Chinese market, and while its open-weight models are available globally to developers, its consumer products are still expanding internationally.

This guide explains Moonshot AI's products, funding, technical innovations, and competitive standing. If you're someone who follows AI developments and also manages digital finances, tools like Bleap can complement your tech-forward lifestyle with 0% FX fees and up to 20% cashback on everyday spending.

Exploring AI from China while spending in euros across borders? Bleap gives you 0% FX fees on every purchase and up to 20% cashback, with no monthly subscription. Get the Bleap card →

1. Moonshot AI: Company Overview and Founding Story

What Is Moonshot AI?

Moonshot AI is an artificial intelligence company based in Beijing, China. It is one of China's 6 "AI Tigers." The company was founded in March 2023 by Yang Zhilin, Zhou Xinyu, and Wu Yuxin, who were schoolfriends at Tsinghua University. It was launched on the 50th anniversary of the release of Yang's favourite album, Pink Floyd's The Dark Side of the Moon, which inspired the name of the company.

Yang's stated goal for founding Moonshot AI is to build foundation models to achieve AGI. Yang's three milestones are long context length, multimodal world model, and a scalable general architecture capable of continuous self-improvement without human input.

Who Founded Moonshot AI? Meet Yang Zhilin

Yang was born in 1992 in Shantou, a city in China's Guangdong province. He studied at Tsinghua University before earning a PhD from Carnegie Mellon University, where he worked under AI researchers Russ Salakhutdinov and William Cohen. During his doctoral studies, Yang interned at Google Brain and Meta while contributing to research on machine learning and large language models.

He has published over twenty papers at top-tier AI conferences such as ICLR, NIPS, ICML, KDD, and ACL. Yang also participated in the research and development of several large models including Google Gemini, Google Bard, PanGu NLP, and Wu Dao. The decision to leave established institutions and return to China to build an independent AI lab signaled a belief that frontier AI could be developed outside Silicon Valley.

2. Kimi: Moonshot AI's Flagship AI Assistant

Core Features and Capabilities

Kimi is an artificial intelligence chatbot and series of large language models developed by Chinese company Moonshot AI. The first version of Kimi supported lossless context of 128,000 characters, making it the first AI model capable of accepting contexts of this size. In March 2024, Moonshot claimed Kimi could handle 2 million Chinese characters per conversation, a significant upgrade from the previous version that could only handle 200,000.

Kimi supports multimodal input including text, documents, images, and web content. It is available via a web app, mobile applications, and API, with both Chinese and English language support. The company sells tiered subscription plans for its chatbot and offers its underlying technology to enterprise clients.

Who Uses Kimi? Real-World Use Cases

Kimi serves a broad range of users. Knowledge workers use it for research summaries, legal document review, and lengthy report analysis, leveraging that long-context capability to process entire books or contracts in a single prompt. Developers rely on it for code generation, debugging, and technical documentation. Students and educators use Kimi for essay writing, study assistance, and curriculum planning. The Kimi chatbot platform has become one of China's highest-profile AI consumer products, while Moonshot AI has increasingly expanded into coding models, agentic AI systems, and enterprise AI services.

3. Kimi Model Lineup: From K2 to K3 and Beyond

Kimi K2 and Kimi K2.6

In July 2025, Moonshot AI released Kimi K2, a 1 trillion parameter mixture of experts large language model with 32 billion active parameters, which was open sourced under a modified MIT license. The Kimi K2 family performs strongly in coding, mathematical reasoning, knowledge Q&A, tool use, and agent workflows.

Moonshot AI released Kimi K2.6 in April 2026, followed by Kimi K2.7 Code in June 2026. Kimi K2.6 is an open-source model featuring state-of-the-art coding, long-horizon execution, and agent swarm capabilities. The company's latest K2-series model, Kimi K2.6, is currently the second-most used LLM on distribution platform OpenRouter.

Kimi K3: What to Expect

Moonshot AI released its Kimi K3 model on 16 July 2026, with 2.8 trillion parameters. It is built on Kimi Delta Attention (KDA), a hybrid linear attention mechanism, and Attention Residuals, with native visual understanding and a 1M-token context window. It is the world's first open-source model in the 3-trillion-parameter class, designed for frontier intelligence scenarios including long-horizon coding, knowledge work, and reasoning.

Upon release, Kimi K3 debuted at No. 3 on the Artificial Analysis AI leaderboard, behind Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol, but outperformed its competitors on Arena.ai's front-end web development benchmark. K3 represents a clear step in Moonshot's long-term scaling strategy, nearly tripling the parameter count from the K2 series.

4. Moonshot AI Funding, Investors, and Financial Backing

Moonshot AI has raised a total funding of $3.77 billion over 4 rounds from 11 investors, including Alibaba Group and Meituan. Moonshot AI's largest funding round so far was a Series D round for $2 billion in May 2026.

Moonshot's backers include Alibaba, Tencent, HongShan (formerly known as Sequoia China), ZhenFund, IDG Capital, and 5Y Capital. In just six months, Moonshot AI's valuation ballooned from $4.3 billion to $20 billion.

Annual recurring revenue crossed $100 million in March 2026, then doubled to over $200 million by April. Capital is allocated toward compute infrastructure, research talent, and expanding Kimi's capabilities. Moonshot AI is now seeking as much as $2 billion in a new funding round that would value the startup at $30 billion.

When you're tracking fast-moving AI companies and their financial trajectories, having your own finances optimized matters too. If you're spending across currencies while researching global tech, Bleap's self-custodial Mastercard gives you 0% FX fees, so you keep more of what you spend.

Managing money across borders while tracking global AI trends? Bleap's savings vaults offer 3.65% AER (Steady) or 3.83% AER (Dynamic) in USD, with $1 minimum deposit and 0% withdrawal fees. No lock-ins. Open a Bleap account →

5. Research and Technical Innovations

The Muon Optimizer

Muon is an optimizer designed for hidden 2D weights of a neural network. It takes gradient of the weight, computes its momentum, and applies Newton-Schulz iterations to orthogonalize the momentum matrix, then uses this orthogonalized matrix to update the weight.

Scaling law experiments indicate that Muon achieves approximately 2x computational efficiency compared to AdamW with compute optimal training. Moonshot open-sourced their distributed Muon implementation that is memory optimal and communication efficient, and also released pretrained, instruction-tuned, and intermediate checkpoints to support future research.

The community reception has been strong. Moonshot AI adopted Muon Optimizer to train its large foundation models like Kimi-K2-Thinking. More recently, Zhipu AI's GLM-5 and DeepSeek-V4 also confirmed the use of Muon Optimizer.

Reinforcement Learning Scaling and PerceptionBench

In their technical report on the Kimi K1.5 model, Moonshot researchers outline their reinforcement learning methods, which they claim enabled the model to achieve state-of-the-art reasoning capabilities on par with OpenAI's o1 model. Moonshot applies reinforcement learning to improve reasoning and agentic reliability across its model lineup. The company has developed internal benchmark suites evaluating multimodal understanding, contributing openly to research discourse in China and globally.

6. Mooncake: Moonshot AI's Serving Infrastructure Platform

Mooncake is the serving platform for Kimi. It features a KVCache-centric disaggregated architecture that separates the prefill and decoding clusters. It also leverages the underutilized CPU, DRAM, and SSD resources of the GPU cluster to implement a disaggregated cache.

Mooncake processes 100 billion tokens daily, which is what makes Kimi's million-context window feasible at scale. Experiments show that Mooncake excels in long-context scenarios, achieving up to a 525% increase in throughput in certain simulated scenarios while adhering to SLOs.

Mooncake won the Best Paper award at FAST 2025, and both the Transfer Engine and Mooncake Store are open-sourced, making the infrastructure relevant for enterprise AI deployment teams worldwide.

7. Agentic AI Capabilities: Kimi as an Autonomous Agent

Agentic AI refers to systems that can plan, use tools, and execute multi-step tasks autonomously, going beyond simple question-and-answer interactions. Kimi's tool-use framework includes web search, code execution, and file parsing.

Moonshot recently launched a general-purpose AI agent, called Kimi Work, built atop its latest K2.6 series models. In November 2025, Moonshot released Kimi K2 Thinking, an open-source update designed for advanced reasoning and agentic tasks. The model features a 1-trillion-parameter MoE architecture with 32 billion active parameters and can execute 200-300 sequential tool calls autonomously.

This capability differentiates Moonshot from pure chat assistants, as Kimi can run automated research pipelines and complete data analysis workflows end-to-end. For developers and businesses building AI-powered automation, Bleap's fee-free crypto trading and self-custodial approach offer a parallel philosophy: tools that put you in control without unnecessary friction or fees.

8. Open-Source Contributions and the Developer Community

Both the code and the model weights for the Kimi K2 series are released under the Modified MIT License. Kimi K2.6 is an open-source model whose weights and code are publicly available on Hugging Face and GitHub, allowing developers and researchers to explore, integrate, and experiment with it.

Moonshot AI has 38 repositories available on GitHub. Open-sourcing models builds trust, attracts global talent, and accelerates community adoption. The modified MIT licence requires attribution only for products exceeding 100 million monthly active users or $20 million in monthly revenue, leaving the models free for almost everyone else to download, fine-tune, and self-host.

Moonshot AI sits within the broader open-source AI ecosystem alongside DeepSeek and Zhipu AI, and community benchmarks from third-party evaluators have consistently validated performance claims.

9. Competitive Positioning in the Global AI Landscape

Moonshot's Kimi competes directly with ByteDance's Doubao, Alibaba's Qwen, Zhipu's Z.ai, and DeepSeek within China. Internationally, Kimi models compete with OpenAI, Anthropic, and Google Gemini, particularly on long-context tasks and agentic capabilities.

Kimi K3 outperforms all rivals except for Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 on overall capability, according to Artificial Analysis. Key differentiators include Moonshot's long-context leadership, infrastructure innovation through Mooncake, and consistent open-weight releases that let the global developer community access frontier-level models.

The Beijing-based company is emerging as one of the best-funded Chinese AI research labs, as investors pour capital into an elite group vying with OpenAI and Anthropic. In the context of US-China AI competition, Moonshot's strategy of releasing open-weight models at competitive price points makes it a company that global enterprises and developers are paying close attention to.

Want to buy and hold crypto with no trading fees while you follow the AI race? Bleap offers fee-free crypto trading, no gas costs, and full self-custody from day one. Use your Bleap Mastercard to spend anywhere, with 0% FX fees and up to 20% cashback. Buy crypto on Bleap →

Frequently Asked Questions About Moonshot AI

What is Moonshot AI and what does it do?

Moonshot AI (月之暗面) is an artificial intelligence company based in Beijing, China. It builds the Kimi AI assistant, a long-context chatbot and API platform, and is known for frontier model research and open-source contributions. It is one of China's 6 "AI Tigers."

Who owns and funds Moonshot AI?

Moonshot's backers include Alibaba, Tencent, HongShan (formerly Sequoia China), ZhenFund, IDG Capital, and 5Y Capital. It was co-founded in 2023 by Yang Zhilin, Zhou Xinyu, and Wu Yuxin. The company has raised a total of $3.77 billion, making it one of China's most-funded AI startups.

What is the Kimi AI assistant and how is it different?

Kimi's first version was known for supporting up to 128,000 tokens of context. It now supports up to 1 million tokens with Kimi K3, multimodal input, and agentic capabilities. It is distinguished by its ability to handle full books, lengthy contracts, and complex autonomous workflows that go far beyond typical chat assistants.

Is Kimi K2 open source?

Yes, Kimi K2.6 is an open-source model. Its weights and code are publicly available on Hugging Face and GitHub, allowing developers and researchers to explore, integrate, and experiment with it. Both the code and the model weights are released under the Modified MIT License.

What is the Muon optimizer developed by Moonshot AI?

Muon (MomentUm Orthogonalized by Newton-schulz) is an optimizer from Moonshot AI that achieves higher sample efficiency compared to AdamW. It applies Newton-Schulz orthogonalization to momentum-based updates. It achieves approximately 2x computational efficiency compared to AdamW, reducing compute costs and training time. The implementation is published openly for the research community.

When was Kimi K3 released?

Moonshot AI released its Kimi K3 model on 16 July 2026, with 2.8 trillion parameters. The full model weights are scheduled for release by July 27, 2026.

Conclusion: Why Moonshot AI Matters

Moonshot AI is a rapidly growing Chinese AI lab built around long-context intelligence, agentic AI, and research-grade innovation. The company released the open-weights Kimi K2 in July 2025 and Kimi K3 in July 2026, demonstrating a clear and aggressive scaling roadmap. Infrastructure like Mooncake (processing 100 billion tokens daily) and research breakthroughs like the Muon optimizer (2x the efficiency of AdamW) give the company a genuine technical edge.

As the global AI race intensifies, Moonshot AI is a company worth watching closely. And if you're someone who stays ahead of technology trends while managing finances across borders, consider pairing your forward-looking mindset with a fintech tool that matches it. Bleap offers a self-custodial Mastercard with 0% FX fees, up to 20% cashback, fee-free crypto trading, and savings vaults at 3.65% or 3.83% AER in USD, all with no monthly subscription. It's a practical way to keep your money working as hard as you do.