Model overview

kimi-k3 appears in the NVIDIA model listing under the publisher Moonshotai. The catalogue describes it as a ~2.8T hybrid KDA+MLA multimodal mixture-of-experts model. Its stated focus areas are long-horizon coding, agentic tool use and image understanding.

  • Publisher: Moonshotai
  • Approximately 2.8T parameters
  • Hybrid KDA + MLA attention
  • Mixture-of-experts architecture
  • Multimodal, including image understanding

Sources: Models | Try NVIDIA NIM APIs, Try NVIDIA NIM APIs

Availability and access

The model catalogue records kimi-k3 as both downloadable and available through a free endpoint. It sits alongside other listed models such as deepseek-v4.1-flash, glm-5-3 and nemotron-3.5-lightning-30b-a3b in the same catalogue. Access is through build.nvidia.com, where models are presented as NIM APIs.

  • Downloadable
  • Free endpoint
  • Listed in the build.nvidia.com models catalogue

Source: Models | Try NVIDIA NIM APIs

Placement on build.nvidia.com

On the build.nvidia.com home page, kimi-k3 is shown under the heading "Use Inference Endpoints" as moonshotai kimi-k3. It is listed with other featured endpoints including deepseek-ai deepseek-v4-pro-0813, nvidia nemotron-3.5-lightning-30b-a3b and nvidia nemotron-3-ultra-550b-a55b. The entry is tagged with descriptors covering mixture-of-experts, multimodal and reasoning characteristics.

  • Featured under "Use Inference Endpoints"
  • Tags include Mixture-of-Experts, Multimodal and Reasoning
  • Shown alongside DeepSeek and NVIDIA Nemotron endpoints

Source: Try NVIDIA NIM APIs

Sources

  1. Try NVIDIA NIM APIs https://build.nvidia.com/ Verified 03 Nov 2026
  2. Models | Try NVIDIA NIM APIs https://build.nvidia.com/models Verified 03 Nov 2026

Last verified 02 Oct 2026. This entry is compiled from the public web pages listed above. Nothing here is stated that those pages do not, and each of them was read on the date shown.