Moonshot AI Unveils Kimi K3, an Open 2.8-Trillion-Parameter Model with 1M-Token Context Window

Moonshot AI has released Kimi K3, what it calls the world's first open 3T-class artificial intelligence model. According to kimi.com, the model is built on new architectural innovations including Kimi Delta Attention and Attention Residuals, and features 2.8 trillion parameters along with a 1-million-token context window and native vision capabilities.

kimi.comKimi K3: World's First Open 3T-Class AI Model with 1M-Token Context Window

"Kimi K3 is the first open 2.8-trillion-parameter model with a 1M-token context window, native vision, and frontier-level agentic coding capabilities."

i-SCOOPKimi K3, the First Open 3T-Class Model

The model is designed for long-horizon coding, knowledge work, and reasoning tasks, positioning it as a competitor in the frontier intelligence space. Despite its open nature and massive scale, kimi.com noted that Kimi K3 still trails the very top proprietary models in overall performance.

"While its overall performance still trails top proprietary models like Claude Fable 5 and GPT 5.6 Sol, it demonstrates frontier-level performance across evaluations, consistently outperforming other tested models."

That distinction highlights the ongoing gap between open and closed AI systems, even as open models grow larger and more capable. The 2.8-trillion-parameter count places Kimi K3 well beyond most open alternatives, yet proprietary rivals maintain an edge in benchmark scores, according to the company's own assessment.

Kimi K3's release underscores the accelerating race to scale open models while pushing context windows to unprecedented lengths, a factor increasingly critical for complex reasoning and multi-step coding tasks. Moonshot AI has made the model openly available, inviting further development and testing from the broader AI community.