NVIDIA has released a quantized Kimi K3.1 Pro model
NVIDIA has released a quantized version of the Kimi K3.1 Pro model from Moonshot AI, optimized for efficient deployment.
Optimization was performed using NVIDIA Model Optimizer and NVFP4 format, which significantly reduces the volume of neural network weights.
Mixture of Experts architecture ensures high performance when processing multimodal data.
- Total number of parameters is 2.8 trillion
- Active parameters reach 104 billion
- The model processes text and images
- Context limit is set at 1,048,576 tokens
The official model card with access to weights is available on Hugging Face at: https://huggingface.co/nvidia/Kimi-K3.1-Pro-NVFP4











