As artificial intelligence evolves from simple answer generators to sophisticated systems capable of deep reasoning and visualization, South Korean startup GPAI is gaining global attention for its impressive performance in one of the world’s most rigorous engineering entrance exams and its innovative tools tailored for STEM education and research.
GPAI, selected by OpenAI as one of South Korea’s leading AI startups, recently claimed that its system scored an equivalent of 351 out of 360 on problems from the JEE Advanced 2025 exam. This benchmark, conducted by the Indian Institutes of Technology (IITs), is renowned for testing profound conceptual understanding and multi-step problem-solving in physics, chemistry, and mathematics. For context, the top human scorer in JEE Advanced 2025 achieved 332/360, while other frontier models like ChatGPT o3 reportedly scored around 327.
The company attributes this performance to a hybrid architecture that decomposes complex problems into smaller logical steps, combining large language model reasoning with a deterministic computational engine. This method not only boosts accuracy but also produces traceable, verifiable solutions—addressing a key demand for explainable AI in academic and technical fields.
Advancing Visual Reasoning in STEM
Building on its reasoning capabilities, GPAI has launched a STEM Visualizer tool that converts text prompts into professional-grade diagrams, technical drawings, schematics, chemical structures, circuit diagrams, and process flows. Users can export these directly to PowerPoint or other formats, edit elements with annotations, arrows, and text, or regenerate sections via additional prompts. The platform incorporates specialized in-house engines for domains like electrical circuits and chemical structures to enhance precision.
GPAI claims to leverage advanced models, such as integrations with GPT Image 2 and others (latest models of GPT, Gemini and Claude), reflecting a flexible strategy in the fast-evolving AI landscape.
The platform further claims to have seen rapid adoption, particularly in India, with reports of over 1,000 users from IIT Delhi joining shortly after launch. It is also expanding in the US, Europe, and across Asia, positioning itself as a go-to AI copilot for students from school level to PhD research, with features like triple-verified solutions, PDF/photo uploads, and cheatsheet builders.
