Moonshot AI Releases Kimi K3 Open-Weight Multimodal Model

TechFlow reports that on July 27, Moonshot AI released the open-weights model Kimi K3 on Hugging Face.

The model is a native multimodal agent model with a total parameter size of 2.8T, adopting an MoE architecture with approximately 104B activated parameters. It supports text and image input and features a 1 million Token context window.

According to the official announcement, Kimi K3 is targeted at long-horizon programming, reasoning, and knowledge work scenarios. It supports deployment via Transformers, vLLM, SGLang, and other methods, with model weights and code released under the Kimi K3 License.