Moonshot AI Launches Kimi-K3, a 2.8T-Parameter Open-Weight Model on HuggingFace

Moonshot AI unveiled Kimi-K3, a 2.8T-parameter open-weight native multimodal agentic model, on HuggingFace on July 27, 2023. This model, built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), is the world's first open 3T-class model, designed for frontier intelligence across long-horizon coding, knowledge work, and reasoning.

Advanced Capabilities

Kimi-K3 is equipped with a 1-million-token context window and native vision capabilities. It advances end-to-end knowledge work, producing deep research with interactive visualizations, widgets, and dashboards, and motion design and video editing.

Accessing Kimi-K3

You can access Kimi K3's API on https://platform.kimi.ai by selecting kimi-k3, and Moonshot AI provides an OpenAI/Anthropic-compatible API. Currently, Kimi K3 is recommended to run on the following inference engines:

• Transformers: Use a pipeline as a high-level helper

  from transformers import pipeline
  pipe = pipeline("image-text-to-text", model="moonshotai/Kimi-K3", trust_remote_code=True)
  messages = [{ "role": "user", "content": [{"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"}] }] 
  pipe(text=messages)
  

• vLLM: Install from pip and serve model

  pip install vllm
  vllm serve "moonshotai/Kimi-K3"
  curl -X POST "http://localhost:8000/v1/chat/completions" 
  -H "Content-Type: application/json" 
  --data '{ "model": "moonshotai/Kimi-K3"}'
  

Technical Specifications

Kimi-K3 is designed to handle complex tasks with its advanced architecture and large parameter count. It is built to support a wide range of applications, from coding assistance to knowledge work and reasoning tasks.

For more details, refer to the Kimi-K3 Technical Report.