News
China's Moonshot AI Releases Kimi K3, Its Largest Open-Weight Model
- By John K. Waters
- 07/17/2026
Chinese artificial intelligence startup Moonshot AI released Kimi K3 this week, an open-weight model with 2.8 trillion total parameters that the Beijing-based company describes as achieving "open frontier intelligence." The release makes K3 the largest open-weight model released to date, exceeding DeepSeek's V4 Pro, at 1.6 trillion parameters, and Zhipu AI's GLM 5 series at 744 billion parameters.
K3 uses a "mixture-of-experts" architecture (MoE), an AI model design that contains many specialized neural networks, called "experts," but activates only a small number of them for each request. It also introduces two techniques developed internally at Moonshot, Kimi Delta Attention (KDA)and Attention Residuals, which the company says improve computing efficiency and reasoning quality. The model supports a 1-million-token context window and processes both text and images natively. Moonshot said K3 is built for long-horizon coding, knowledge work, and reasoning tasks that require sustained, multistep execution with limited human oversight.
An independent evaluation by Artificial Analysis, a third-party AI benchmarking firm, gives K3 a score of 57 on its Intelligence Index, ranking it fourth among 189 models tested. That places K3 behind Claude Fable 5 and two configurations of OpenAI's GPT-5.6 Sol, and ahead of Anthropic's Claude Opus 4.8, OpenAI's GPT-5.5, Claude Sonnet 5, and GLM-5.2. Moonshot's own launch materials report additional coding benchmark results, including a first-place finish on two internal coding tests, though those figures have not yet been independently reproduced using a common testing harness.
In its release announcement, Moonshot said, "K3 stands as Moonshot AI's most powerful open-source coding model to date." The company also said K3 performed competitively against Claude Fable 5, the most advanced model widely available from Anthropic, while outperforming Claude Opus 4.8 and OpenAI's GPT-5.6 Sol and GPT-5.5 on its benchmark set.
Kimi K3's API is live now through Moonshot's platform, the Kimi consumer app, Kimi Work, and the Kimi Code coding tool, with additional access available through OpenRouter. The API is compatible with the OpenAI software development kit, which lowers the integration barrier for developers already building on OpenAI or Anthropic tools. API pricing is set at $3 per million input tokens and $15 per million output tokens, the highest rate among Chinese AI labs currently, though still below the per-task cost of Anthropic's Opus 4.8, according to Moonshot. Full model weights are scheduled for release on July 27.
Moonshot AI was founded in 2023 by Tsinghua University graduate Yang Zhilin and is backed by Alibaba and Tencent. Its Kimi chatbot ranked third among Chinese AI products by monthly active users before DeepSeek's R1 model disrupted the domestic market in January 2025, after which Kimi's ranking fell to seventh. Moonshot's shift toward open-weight releases began with Kimi K2 in July 2025 and accelerated with K2.5 in January 2026, a strategy the company has used to rebuild its position in China's AI market.
Moonshot's approach reflects a broader pattern among Chinese AI developers, including DeepSeek, Alibaba, Tencent, and Baidu, all of which have released open-weight models over the past year. Analysts have offered several explanations for the strategy. Bloomberg has reported that Chinese firms, facing U.S. restrictions on access to the most advanced AI chips and unable to match the capital spending of American rivals, are using open-weight releases to speed AI adoption across their domestic economy. Reuters has reported that open-sourcing allows Chinese companies to demonstrate technical capability and expand their developer communities and global influence, a strategy analysts say could help China counter U.S. efforts to limit its technological progress. Separately, MIT Technology Review has reported that Chinese firms increasingly view open-weight releases as the fastest way to close the capability gap with U.S. labs by attracting developers, driving adoption and helping set technical standards.
The release came one day ahead of the 2026 World Artificial Intelligence Conference in Shanghai, which opened July 17 with a keynote address from Chinese President Xi Jinping.
About the Author
John K. Waters is the editor in chief of a number of Converge360.com sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS. He can be reached at [email protected].