Moonshot AI has released the full model weights and technical report for Kimi K3, opening its most capable artificial intelligence model to researchers and developers.

Releasing the model weights and technical report of Kimi K3.

Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window.

New model architecture: 2.5x the intelligence per unit of compute, not just more params.

Alongside… pic.twitter.com/Yz5uWeMbIm

— Kimi.ai (@Kimi_Moonshot) July 27, 2026

The company published the weights on Hugging Face and released the technical report through its official GitHub repository. Both the code and model weights are available under the Kimi K3 License.

Moonshot initially introduced Kimi K3 earlier this month but said the complete weights and additional technical details would arrive by July 27. The latest release completes that rollout.

Kimi K3 has 2.8 trillion total parameters and activates 104 billion parameters during inference. It uses a mixture of experts architecture with 896 experts, selecting 16 for each token, and supports a context window of more than 1 million tokens.

The report also provides additional information on Kimi Delta Attention, Attention Residuals and the model’s native quantization system. Moonshot said the architectural changes deliver approximately 2.5 times greater scaling efficiency than Kimi K2.

The release allows developers to download and deploy the model through frameworks including Transformers, vLLM and SGLang. Moonshot recommends large supernode configurations due to the model’s size and infrastructure requirements.