Moonshot AI Open-Sources Kimi K3.1 Pro Weights

November 5 — Moonshot AI has released the open weights of Kimi K3.1 Pro on Hugging Face. According to the model card, Kimi K3.1 Pro is a native multimodal agent model built on a MoE architecture, with 2.8T total parameters and 104B activated parameters, a context length of about 1 million tokens, and is the first model in the K3.1 series; the weights are released in MXFP4 quantized form (activations in MXFP8, using quantization-aware training). The official API is also open, and developers can select kimi-k3.1-pro on platform.kimi.ai for inference, compatible with OpenAI and Anthropic interfaces. Previously, Moonshot AI released Kimi K3.1 Pro on October 24 and said the model weights would be open-sourced before November 4.