Moonshot AI has released the open weights of Kimi K3.1 Pro on Hugging Face, fulfilling its promise to make the model available before November 4. The native multimodal agent model features a Mixture-of-Experts architecture with 2.8 trillion total parameters and 104 billion activated parameters, supporting a context length of approximately one million tokens. The weights are distributed in MXFP4 quantized form, with activations in MXFP8 using quantization-aware training. Moonshot AI also opened its official API, allowing developers to access kimi-k3.1-pro via platform.kimi.ai with compatibility for both OpenAI and Anthropic interfaces.