Solar Pro 2 (Reasoning) is a 31-billion parameter language model developed by Upstage AI, designed to provide advanced logical deduction and multi-step task execution. As a successor to the initial Solar Pro series, it introduces a hybrid architecture that allows users to switch between a standard Chat Mode and a specialized Reasoning Mode. This mode utilizes Chain-of-Thought (CoT) processing to enhance performance in analytical domains such as mathematics, coding, and structured problem-solving.
The model is engineered to deliver performance comparable to much larger frontier models while maintaining the efficiency required for single-GPU deployment. It supports a context window of 65,536 tokens, enabling the analysis of large documents and complex, multi-turn instructional prompts. Solar Pro 2 is particularly recognized for its multilingual proficiency, specifically optimized for high accuracy in Korean, Japanese, and English benchmarks.
In addition to its reasoning capabilities, Solar Pro 2 serves as an agentic foundation model, capable of interacting with external tools and executing complex workflows. It has been evaluated favorably on global benchmarks like MMLU-Pro and Arena-Hard, where its performance-to-size ratio positions it as a highly efficient alternative to frontier-scale systems.