As recorded at collection: NVIDIA NIM Kimi K3 full-performance model, free
As recorded at collection: A post says that the full-performance Kimi K3 model is available for free through the AI model API service NVIDIA NIM. There is not yet a model page on the official website, and only the API endpoint is working.
- What you get
- Free use of the full-performance Kimi K3 model
- Price
- Free
- Valid until
- To be confirmed
- Requirements
- No model page on the official website yet · Only the API endpoint is working
Details
The full-performance Kimi K3 model offered for free through the NVIDIA NIM API has a scale of 2.8T. Its context length per request is 1 million, and it supports image recognition and three thinking modes: low, high, and max.
There is not yet a model page on the official website, but the API endpoint is working. It was tested on OMP. The specific access address and signup and authentication methods need to be confirmed.
In the OMP test, the response-speed metric, throughput in tokens per second (tps), was measured at about 50. The assessment was that it was fast enough, gave smart responses, and had no interruptions. Availability by region, signup requirements, usage limits, deadline, and payment terms need to be confirmed.
Sources and verification history
- Collection channel
- NodeLoc
- Collected at
- August 21, 2026, 11:44 KST
- First published
- Updated