South Korean AI startup Upstage has made significant progress in the global artificial intelligence race with its latest large language model Solar Pro 2, which has outperformed OpenAI's GPT-4.1 in the prestigious Artificial Analysis Intelligence Index. Released on July 21, 2025, Solar Pro 2 scored 58 points in the Intelligence Index — the highest among all Korean-developed LLMs and five points above GPT-4.1.
The achievement puts Korea alongside AI giants like OpenAI, Google, and Meta in the top 10 of the world's most advanced frontier model developers, representing a major milestone for the country's AI ecosystem. The Artificial Analysis Intelligence Index evaluates models across seven key benchmarks, including reasoning, general knowledge, mathematics, and coding.
What makes Solar Pro 2 particularly noteworthy for developers is its efficiency. With just 31B parameters, it delivers top-tier performance through world-class multilingual support, advanced reasoning, and real-world tool use. In benchmark tests such as MMLU-Pro, Math500, AIME, and SWE-Bench, Solar Pro 2 delivered results comparable to leading frontier-scale models including OpenAI's GPT-4o, DeepSeek's R1, Mistral's Small 3.2, and Alibaba's Qwen 3—despite being less than half their size.
The model shows particular strength in Korean language processing. Across key benchmarks like Arena-Hard-Auto, Hae-Rae, and Ko-MMLU, the model matched or outperformed language models by tech giants. Beyond Korean, it demonstrates impressive performance in major language benchmarks for English and Japanese, breaking down global communication barriers.
Solar Pro 2 has also successfully surpassed Chinese models like DeepSeek V3 (53 points) and Moonshot AI's Kimi K2 (57.59 points). In addition to outperforming larger models in reasoning-heavy tasks, it also ranked highly in cost-efficiency metrics, making it a viable option for enterprise use where performance and cost must be carefully balanced.
Looking ahead, Upstage is currently developing a 100B+ parameter model aimed at pushing further into the upper tier of the Intelligence Index. This development could further strengthen South Korea's position in the competitive global AI landscape, where efficiency and performance increasingly matter more than model size alone.