Moonshot AI

Overview

Moonshot AI is a Chinese artificial intelligence company founded in March 2023 by Yang Zhilin, Zhou Xinyu, and Wu Yuxin, all graduates of Tsinghua University. Yang Zhilin, the CEO, graduated first in his class from Tsinghua's computer science program, completed his PhD at Carnegie Mellon University in under four years, and co-authored the Transformer-XL and XLNet papers, both foundational works in modern language modeling. He previously worked at Meta AI and Google Brain. The company is one of China's "AI Tigers," a group of leading Chinese AI startups that also includes DeepSeek and Zhipu AI. In January 2026, Moonshot AI closed a $500 million Series C round led by IDG Capital, with participation from Alibaba and Tencent, valuing the company at $4.3 billion post-money. The company holds cash reserves exceeding 10 billion yuan (approximately $1.4 billion USD). Moonshot AI's flagship model is K2.5, released in January 2026, a roughly 1 trillion parameter Mixture of Experts model with approximately 32 billion active parameters per forward pass, 384 total experts (8 selected per token plus 1 shared expert), and a 256K token context window. K2.5 competes head to head with GPT-5.2, Claude Opus 4.6, and Gemini 3.1 Pro on major benchmarks and is released under a Modified MIT License. The company's product lineup includes Kimi (the AI chat assistant), Kimi Code (an open source CLI coding agent), the Kimi API Platform (developer access to Moonshot models), and Agent Swarm (a multi-agent orchestration system that can coordinate up to 100 parallel sub-agents). K2.5 weights are hosted on Hugging Face and the model is available through US-based inference providers including Fireworks, OpenRouter, Together AI, and DeepInfra.

What makes them different

Compared to OpenAI and Anthropic, Moonshot AI delivers frontier level model performance as open source. K2.5 matches GPT-5.2 and Claude Opus 4.6 on major benchmarks while its weights are freely available under a Modified MIT License. Annual cost estimates for equivalent workloads put K2.5 at approximately $13,800 compared to $56,500 for GPT-5.2 and $150,000 for Claude Opus 4.6. The trade off is a less polished consumer experience and a smaller ecosystem of integrations. Compared to DeepSeek, both companies are Chinese labs releasing frontier open source models, but Moonshot AI differentiates through its Agent Swarm system and stronger multimodal vision capabilities. K2.5 scored highest on 9 out of 17 image and video benchmarks when compared against GPT-5.2, Claude Opus 4.6, and Gemini 3.1 Pro. DeepSeek holds the advantage on raw cost efficiency at the API level. Compared to Google's Gemini, Moonshot AI cannot match Google's ecosystem breadth, but K2.5's open source availability means organizations can self-host the model and avoid per-token API costs entirely. Google's models remain proprietary. Unique strengths include the Agent Swarm system, which uses Parallel-Agent Reinforcement Learning (PARL) to coordinate up to 100 parallel sub-agents across 1,500 steps without predefined roles or hand-crafted workflows. The company also leads in multimodal vision with MoonViT, a custom 400 million parameter vision encoder for native image and video understanding. K2.5 delivers frontier performance at approximately 76% lower cost than Western competitors, and the model is globally accessible through both direct API and US-based inference providers.

Their tools

Updated

Kimi

moonshot-ai

chatbot

Powerful AI assistant with Agent Swarm, 256K context, multimodal vision, visual coding, and plans from free to $199/mo. Built by the fastest decacorn in AI.

FreeView details →

Subscription plans

Links