.avif)
South Korean AI pioneer Upstage has released Solar Pro 2, its cutting-edge large language model (LLM), to the public. Designed with just 31 billion parameters, Solar Pro 2 matches—and at times surpasses—powerhouses like Llama 3.3 70B and Qwen2-72B in reasoning, agility, and multilingual performance.
Solar Pro 2 particularly stands out in the Korean language, rivaling major models such as GPT-4 and Claude 3 across key benchmarks like Ko-MMLU, Hae-Rae, and Ko-IFEval. Businesses operating in finance, healthcare, and law can benefit from its advanced accuracy and deep comprehension, not only in Korean but also in English and Japanese.
A key innovation is its "Reasoning Mode," allowing for advanced multi-step problem solving, complex mathematical operations, and logical analysis. Solar Pro 2 is also optimized for rapid tool integrations and software connections, instantly processing documents and extracting structured information from them. These features position it as a game-changer for efficiency in sectors like finance, healthcare, insurance, and manufacturing.
What sets Solar Pro 2 apart is not just its intelligence but also its compact footprint and hardware efficiency, offering high performance with lower computing requirements—leading to substantial cost savings for organizations. Solar Pro 2 is available for free testing through the Upstage Console until July 15, 2025.
With this launch, Upstage signals its ongoing commitment to AI innovation and has announced plans to showcase real-world use cases in industries such as pharmaceuticals, insurance, finance, and manufacturing in the near future.