Deep Sik V3.1 was China’s cutting-edge model at the time, released around the end of August, and a slightly better version was released in Korea at the end of December. It only took about 4 months to catch up with China’s leading edge.
2. 235B LG K-ExaOne
Ahead of the QWEN3 235B that came out at the end of July 2025 (not by much, but slightly) (Alibaba’s top model at the time)
https://biz.chosun.com/industry/company/2025/12/30/MLDD4KEPTFHGPGBKJBMMXS7VNY/
3.100B Upstage Solar Open has been released with performance equivalent to GPT OSS 120B.
(GPT OSS 120B has GPT-o4 mini-high performance)
If this is the case, Korea also has potential. There is no country other than the United States and China where five models of this size are released at once.
The only one that has received attention overseas is the Mistral, but their performance is not that impressive these days. I was pushed back. If this continues, we might be able to make it to the top three, excluding the US and China.
With 1000 GPUs, we can keep up with China with a 4 month gap.
Is it really possible for Sovereign AI LLM?
-5