2.8T total / 104B active (MoE)
1,048,576 tokens
Kimi K3.1-Pro License
Open-weight, natively multimodal agentic flagship and first release in Moonshot AI's K3.1 series, post-trained from Kimi K3 using substantially greater post-training compute and large-scale joint reinforcement learning across hundreds of downstream domains. The 93-layer Mixture-of-Experts architecture combines 69 Kimi Delta Attention layers with 24 Gated DeepSeek Sparse Attention layers, replacing K3's full-attention Gated MLA layers for more efficient long-context modeling. It uses one initial dense layer, 896 routed experts with sixteen selected per token plus two shared experts, a 7,168-dimensional attention hidden state, 96 attention heads, a sparse-attention index top-k of 2,048, and the SiTU-GLU activation function. Native multimodal support covers text and images through a 401M-parameter MoonViT-V2 vision encoder. The released checkpoint was quantization-aware trained with MXFP4 weights and MXFP8 activations; its official Hugging Face repository contains 96 downloadable safetensor shards totaling approximately 1.561 TB. K3.1 Pro supports low, high and max reasoning effort, defaults to max, and requires complete reasoning content and tool-call history to be preserved across multi-turn conversations. Official deployment guidance covers vLLM, SGLang and TokenSpeed. Moonshot emphasizes long-horizon coding, full-project game development in Unreal Engine, Unity and Godot, parametric 3D creation in Blender and CAD, financial research, factual accuracy and other agentic knowledge work. Publisher-reported results include 65.7% on Terminal-Bench 4.0, 61.2% on Kimi Code Bench 3.0 Long-horizon, 72.8 on BenchCAD without tools, 92.7 with tools, and 64.4 on OfficeQA Pro v2. On the internal Kimi Hallu Bench Hard evaluation, Moonshot reports that claim-level hallucination fell from 23.4% for K3 to 8.3% for K3.1 Pro. The released cyber-controlled checkpoint scores 81.0% pass@3 on Moonshot's internal xdev-bench-hard, compared with 90.5% for an unrestricted experimental checkpoint; Moonshot says it removed cyber-related data during mid-training and filtered cyber reinforcement-learning environments before release. The custom Kimi K3.1-Pro License broadly permits use, modification, distribution, fine-tuning and sale, but model-as-a-service businesses exceeding $20 million in aggregate revenue over any consecutive twelve months require a separate Moonshot agreement. Commercial products exceeding 100 million monthly active users or $20 million in monthly revenue must prominently display the Kimi K3.1-Pro name; internal use is exempt from those additional requirements.