The original title is: "Overview of Kimi-K3: Moonshot AI's 2.8T parameter open-weight multimodal model"

Short summary

Kimi-K3 by Moonshot AI is a 2.8-trillion parameter open-weight multimodal model with a 1-million-token context window, native text/image/video processing, and a Stable LatentMoE architecture activating 16 of 896 experts per token. It targets long-horizon coding, agentic knowledge work, vision-in-the-loop design, and financial/legal document analysis. Benchmark scores include 81.2 on FrontierSWE and 88.3 on Terminal-Bench 2.1 for agentic coding tasks.

• •2.8T parameter open-weight multimodal model with 1M token context window
• •Excels at long-horizon coding, vision-in-loop design, and financial/legal document analysis
• •Uses Kimi Delta Attention and LatentMoE architecture with MXFP4 quantization-aware training

Generated with AI, which can make mistakes.