Training and Researching to enable the Frontier LLM experience on Local AI
Swift, our open models
Swift uses 58.5% fewer reasoning tokens while keeping the accuracy of its answers.
6.1M+downloads
#1trending finetune on Hugging Face in launch week
#9trending model overall on Hugging Face in launch week
GSQ-RCO quantization with the help of

Used by engineers and researchers at



Meet Ukis Foundation Models
Releasing Nov. 2026
- 1BPhoneOpen-source personal agents on any phone
- 4BCarIn-car assistant while being offline
- 8BLaptopCoding agent on a GPU-less laptop
- 27B-A3BComputerComputer-use agents on any desktop
