Moonshot AI Releases Kimi K3 Open-Weight Multimodal Agentic Model
Moonshot AI’s Kimi K3 model card announces a 2.8T-parameter Mixture-of-Experts model with 104B activated parameters, native vision capabilities, and a 1,048,576-token context window. The model uses Kimi Delta Attention and Attention Residuals, while its full weights are released under the Kimi K3 License and its API supports OpenAI- and Anthropic-compatible interfaces. Moonshot AI also presents benchmark results and a 2.5× scaling-efficiency improvement over Kimi K2, but the supplied report does not independently establish those vendor claims.
Kimi K3 combines open weights, native multimodality, agentic coding, long context, and API availability at a very large model scale, making it a notable model release for AI practitioners.
moonshotai/Kimi-K3 Image-Text-to-Text • 2.8T • Updated 1 day ago • 735k • • 12k
tags:
- compressed-tensors
- conversational license: other license_name: "kimi-k3" library_name: transformers pipeline_tag: image-text-to-text new_version: moonshotai/Kimi-K3.1-Pro

Some source content is omitted. Continue reading on the original site.
#Moonshot AI#Kimi K3#Kimi Delta Attention#Attention Residuals#Kimi Code CLI#Kimi K3 License#MoonViT-V2