Registered as a Global Frontier Model by Artificial Analysis... First in Korea
Outperformed Big Tech LLMs such as GPT-4.1 and Llama Maverick
Achieved Comparable Performance with Fewer Parameters than Global Competitors
On July 21, Upstage announced that its self-developed next-generation large language model (LLM), Solapro 2, has been selected as the only Korean frontier model in a global artificial intelligence (AI) model performance evaluation.
This evaluation was based on the Intelligence Index recently published by the independent LLM performance analysis agency Artificial Analysis. The index assesses top global models using seven benchmarks, including reasoning, general knowledge, mathematics, and coding.
Artificial Analysis released the intelligence index. Upstage's Solapro 2 model ranked 12th. Provided by Upstage
In this announcement, Solapro 2 was registered as the only Korean model on the Artificial Analysis platform, which selects only the highest-performing models. This is the first time a domestically developed LLM has been recognized as a global frontier-level model. Upstage was listed as one of the top 10 frontier model developers alongside global big tech companies such as OpenAI, Google, and Meta.
Solapro 2 scored 58 points, surpassing big tech LLMs such as Mistral Small (55 points, Mistral), GPT-4.1 (53, OpenAI), Llama 4 Maverick (51, Meta), and GPT-4o (41, OpenAI). It also outperformed Chinese models such as DeepSeek V3 (53) and Moonshot AI's Kimi K2 (57.59).
The model also demonstrated high efficiency in terms of "cost-effectiveness." The major models are large-scale, equipped with 10 to 20 billion parameters, while xAI's Grok 4, which recorded the highest performance, has as many as 1.7 trillion (1.7T) parameters. In contrast, Solapro 2 has only 3.1 billion parameters. Despite having only about 15% of the parameters compared to major global models, it achieved a similar level of performance.
Performance-to-cost table published by Artificial Analysis. Upstage's Solra Pro 2 model was evaluated as delivering decent performance while maintaining low operating costs. Provided by Upstage
Elon Musk, CEO of Tesla, also showed interest in the results. He founded the AI startup xAI, which has released the artificial intelligence service "Grok." He reposted the announcement on his X account, stating, "But xAI's Grok is still number one. And Grok is improving rapidly." While xAI's Grok 4 was evaluated as having the highest performance in this evaluation, it ranked in the lower to mid-tier in terms of operating costs and cost-effectiveness indicators.
Upstage plans to develop a next-generation LLM with 100 billion parameters as a follow-up model.
Kim Sunghoon, CEO of Upstage, said, "We are pleased to prove that our foundation model Solapro 2, developed with domestic technology, can stand shoulder to shoulder with the world's top models," adding, "We will continue to further advance our proprietary models with independent technology, contributing to South Korea not only joining the global AI top three but also taking the lead in the global technology market."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
