Press Release
June 30, 2026
Nota AI, an AI model compression and optimization company led by CEO Myungsu Chae, announced that it has successfully optimized LG AI Research’s flagship AI model, K-EXAONE 236B, on Furiosa AI’s data center neural processing unit (NPU).
K-EXAONE 236B is a large-scale AI model with approximately 236 billion parameters and is built on a Mixture-of-Experts(MoE) architecture, which selectively activates multiple expert models. While MoE structures can improve the efficiency of large models, they require highly sophisticated optimization to ensure that each expert model operates reliably.In particular, frontier-scale models often go through long reasoning processes when solving complex problems, meaning even small errors introduced during quantization can accumulate and affect the accuracy of the final output. This achievement is significant in that Nota AI optimized such a large model for a Korean NPU environment while maintaining accuracy across major evaluations.
In this project, Nota AI optimized K-EXAONE for Furiosa AI’s data center NPU. Rather than readjusting the entire model, Nota AI precisely analyzed specific sections where performance degradation could occur and applied optimization only where necessary to minimize performance loss. As a result, the company enabled efficient execution of a large AI model on a Korean NPU while maintaining performance at a level comparable to the base line model across key metrics.
The performance evaluation also showed meaningful results. Nota AI reduced the size of K-EXAONE by approximately 71%,lowering the memory burden required to run a large AI model, while maintaining accuracy close to the original model across key evaluation categories,including scientific reasoning, instruction following, and mathematical problem solving. This demonstrates the potential to run a 236-billion-parameter model more efficiently and improve the operational efficiency of data center AI infrastructure.
In Nota AI’s internal evaluation environment, the optimized model recorded 79.80 on GPQA for scientific reasoning, 68.98 on IFBench for instruction following, and 88.57 on AIME25 for mathematical problem solving. Before optimization, the original model scored 79.1, 67.3, and 92.8, respectively. Based on the simple average of the three major evaluation categories, the optimized model maintained approximately 99.2%of the original model’s accuracy.
This achievement goes beyond simply running a large AI model on a Korean NPU. It confirms that performance and stability required for real-world services can be maintained in such an environment. In particular, the result is seen as a meaningful example of how Furiosa AI’s data center NPU, LG AI Research’s advanced AI capabilities, and Nota AI’s model optimization technology can work together to support the operation of high-performance LLMs within Korea’s AI ecosystem.
Recently, access to cutting-edge AI models and the infrastructure required to run them has become an important issue in the global AI industry. Following discussions around export controls on certain AI models and infrastructure, countries are increasingly focusing on securing their own AI models and computing infrastructure as part of the sovereign AI movement. In this context, Nota AI’s achievement highlights the need for domestic AI semiconductors, domestic AI models, and the optimization technologies that connect them to advance together.
“As sovereign AI gains increasing attention, what matters most is connecting models, semiconductors, and optimization software into an executable AI infrastructure,” said Myungsu Chae,CEO of Nota AI. “This achievement demonstrates the real-world operational potential of large AI models by combining Furiosa AI’s data center NPU, LG’s flagship AI model K-EXAONE, and Nota AI’s optimization technology.”
