Korean AI startup Upstage has made history. Its proprietary large language model (LLM), Solar Pro 2, has been named a global AI frontier model after outperforming top-tier models, including OpenAI’s GPT, in the latest Intelligence Index rankings. This makes Solar Pro 2 the only Korean model recognized among the world’s leading AI frontier models—a significant milestone for the domestic AI ecosystem.
Upstage ‘Solar Pro 2’ Selected as Global AI Frontier Model
Korean AI startup company Upstage announced that its next-generation large language model (LLM), Solar Pro 2, had become one of top performing global AI frontier models.
On July 21, 2025, Artificial Analysis, an independent LLM performance assessment platform known for exclusively featuring top-tier AI models officially listed ‘Solar Pro 2’ on its Intelligence Index. Solar Pro 2 scored an impressive 58 points, the highest among Korean-developed LLMs.
Artificial Analysis’s Intelligence Index assesses the world’s leading AI models using seven core benchmarks, including reasoning, general knowledge, mathematics, and coding. Thanks to its high transparency and objectivity, the index is widely trusted in the industry and frequently cited by big tech companies.
‘Solar Pro 2’: The Only Korean AI Frontier Model, Beating GPT by 5 Points
With this recognition, Upstage joins the ranks of global tech giants like OpenAI, Google, and Meta, as the only Korean AI model among the top 10 frontier model developers worldwide.
And as the platform successfully scored 58 points, Solar Pro 2 has successfully outperformed several major big tech LLMs, such as:
- Mistral Small (55 points)
- GPT-4.1 (53 points)
- Llama 4 Maverick (51 points)
- GPT-4o (41 points)
Not only that, but it has also successfully surpassed Chinese models like DeepSeek V3 (53 points) and Moonshot AI’s Kimi K2 (57.59 points).
High Efficiency and Cost-Effective
In addition to being one of the top performing AI frontier models, Solar Pro 2 is also considered one of the most efficient and cost-effective.
Most of these global AI frontier models operate with massive 100 billion to 200 billion parameters. The highest-performing model, Grok-4 by xAI, even runs on an astounding 1.7 trillion parameters.
But not Solar Pro 2. The Korean AI model successfully operates within only 31 billion parameters, but it is still capable of delivering performances on par with these much larger models.
Hence, it highlights Upstage’s exceptional technological competitiveness and efficiency in small language models.
As in direct cost-performance comparisons, Solar Pro 2 outperforms GPT-4.1, which has been widely used in enterprise environments. It also delivers greater efficiency than Claude 4 Sonnet and Mistral Small, making it an attractive option for businesses prioritizing both performance and operational cost savings.
Bringing South Korea to the Highest Tier of Global AI Frontier Index
Finally, Upstage also declared that the company would be developing its own next-generation LLM with 100 billion parameters to bring South Korea to the top of global AI leaders.
“We are proud that Solar Pro 2, a foundation model developed entirely with Korean technology, stands shoulder to shoulder with the world’s top models. We will continue to refine our models using our own proprietary technology, contributing to Korea not only being a global AI powerhouse but also securing a leadership position in the international AI market.”
Kim Sung Hoon, CEO of Upstage.
Keep tab on latest news in the Korean startup ecosystem & follow us on LinkedIN, Facebook, and Twitter for more exciting updates and insights.
