Kimi K3.1 Pro is an open-weight, natively multimodal agentic model, and our most capable 3T-class model to date. Post-trained from Kimi K3 with large-scale joint reinforcement learning across hundreds of downstream domains, it is the first model in the K3.1 series. With a 1-million-token context window, Kimi K3.1 Pro delivers substantial improvements in coding, 3D creation, financial research, factual accuracy, and other agentic capabilities.

Key Features

• Efficient Long-Context Architecture: Kimi K3.1 Pro keeps Kimi K3’s 2.8T-parameter MoE design, activating 16 of 896 experts (104B active parameters), and replaces full-attention layers with Gated DeepSeek Sparse Attention for more efficient long-context modeling.
• Long-Horizon Coding: Kimi K3.1 Pro sustains long engineering sessions, from quick fixes to full-project builds, with stronger game development and script-driven 3D modeling in tools like Blender, CAD software and Unity.
• Financial Research & Factual Accuracy: Kimi K3.1 Pro handles financial modeling, research and business operations, with improved factual accuracy over Kimi K3.
• Native Multimodality & Long Context: Kimi K3.1 Pro understands text and images, and supports a 1-million-token context window.
• Open Frontier Weights: We release the full Kimi K3.1 Pro model weights under the Kimi K3.1 Pro License.