Moonshot AI has released the full model weights and technical report for Kimi K3, making its 2.8 trillion parameter AI model available to developers and researchers. The model uses a mixture of experts architecture with 896 experts, selecting 16 for each token, supporting a context window exceeding 1 million tokens. The company states these architectural changes deliver approximately 2.5 times greater scaling efficiency than Kimi K2. Both the weights and code are available on Hugging Face and GitHub under the Kimi K3 License, through the Transformers, vLLM and SGLang frameworks.
Source: Read the original article
