Reasoning AI agent developer Turing said on Tuesday it is supplying high-quality math data to major companies participating in the government’s Indigenous AI Foundation Model Project (Dokpamo).

Turing is providing more than 300,000 selected questions as AI training data from those accumulated through its math services, Math King and GPAI. The dataset covers the full range of elementary, middle and high school curricula and is organised by difficulty. It also includes questions with images such as diagrams and graphs, supporting multimodal AI training.

Demand for math data has been rising as logical reasoning capability emerges as a key competitive factor for large language models (LLMs), but existing open-source data has faced constraints on commercialisation due to errors and copyright issues. Turing has signed supply contracts with major AI companies, including Dokpamo participants, based on its own data that overcomes these limits, and has secured multiple renewals, the company said.

The data is supplied through two methods, rental and purchase, and can be customised depending on client needs, including difficulty composition and whether to include images. It is delivered with data structuring, preprocessing and format conversion, including JSON and LaTeX, and an English version can also be supplied.

Turing CEO Minkyu Choi (최민규) said, "To strengthen AI reasoning capabilities, we are providing math data in a form that can be used immediately for training and evaluation." He added, "We will supply it flexibly and quickly to match companies’ purposes and formats."