Kimi-K3

About

What Kimi-K3 is

Model Overview Description: The NVIDIA Kimi-K3-NVFP4 model is the quantized version of Moonshot AI's Kimi-K3 model, which is an auto-regressive language model that uses an optimized transformer architecture. Kimi-K3 is a flagship, natively multimodal Mixture-of-Experts (MoE) model designed for long-horizon coding, end-to-end knowledge work, game development, deep reasoning, and visual-agent use cases. It has 2.8 trillion total parameters and a native 1,048,576-token context window.For more information, please check here. The NVIDIA Kimi-K3 NVFP4 model is quantized with Model Optimizer. This model is ready for commercial or non-commercial use. References - Nvidia Model Optimizer: https://github.com/NVIDIA/Model-Optimizer - Kimi-K3 model card: https://huggingface.co/moonshotai/Kimi-K3 License/Terms of Use: Governing Terms: Use of this model is governed by the NVIDIA Open Model Agreement. Additional Information: Kimi K3 License. Kimi K3. Deployment Geography: Global Use Case: Kimi-K3 use case: developers and inference providers who need ready-to-deploy, pre-quantized multimodal agentic models for NVIDIA GPU inference, including long-horizon coding, visual-agent workflows, game-development tasks, and end-to-end knowledge work. Release Date: Hugging Face 08/14/2026 via https://huggingface.co/nvidia/Kimi-K3-NVFP4 Model Architecture: Architecture Type: Transformers Network Architecture: Mixture-of-Experts transformer using Kimi Delta Attention (KDA), Attention Residuals (AttnRes), and Stable LatentMoE Number of Model Parameters: 2.8T total parameters; 16 of 896 experts are activated per token Input: Input Type(s): Text, Image, Video Input Format(s): String, Binary(Base64 encoded), Binary(Base64 encoded) Input Parameters: One-Dimensional (1D), Two-Dimensional (2D), Three-Dimensional (3D) Other Properties Related to Input: Native context length: 1,048,576 tokens. Images may be supplied as base64-encoded data, and video-file inputs are supported. Output: Output Type(s): Text Output Format: String Output Parameters: 1D (One-Dimensional): Sequences Other Properties Related to Output: Outputs may include natural-language responses, code, structured JSON, tool-call requests,

Summarised from the published model card. Read the full card on the HuggingFace links below.

Specifications

What it is

How it works

How language models work

Variants

Sizes & precisions

Open weights ship in several sizes and precisions. One page, all the variants — pick the one that fits your GPU. VRAM figures are estimates from model size.

Using it via the API

Call it like any OpenAI endpoint

Once AxForge deploys nvidia-kimi-k3 for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (nvidia-kimi-k3 below is illustrative; you get the exact model name on deployment.)

$ curl -sS https://api.axforge.ai/v1/chat/completions \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"nvidia-kimi-k3","messages":[{"role":"user","content":"Hello"}]}'

Details

Languages, data & research

Licence

Commercial licence needed

Sources

Weights & code

Explore

More language models