LG AI Research Institute opens Dokpamo ‘K-ExaOne 2.5’ API for free

‘K-ExaOne 2.5’, an independent AI foundation model (hereinafter referred to as Dokpamo) developed by LG AI Research Institute, is available for free.

LG AI Research Institute participated in 'NVIDIA AI Day Seoul 2026' held at COEX in Seoul on the 9th and announced that it would open the API (Application Program Interface) of 'K-ExaOne 2.5' so that anyone can use it.

The API can be downloaded from the K-Exaone platform (docs.exaone.ai), and the plan is to expand opportunities to directly use Korean AI models and apply them to services.

'K-ExaOne 2.5' is a technology verification model that applied additional learning to the 2nd DokpaMo released in August.

LG AI Research Institute announced that, as a result of its self-evaluation, 'K-ExaOne 2.5' showed performance comparable to the world's top weighted open model in the field of agentic AI, where AI performs several steps on its own.

At the same time, the results of four performance indicators that evaluate agentic AI capabilities were revealed. The comparison targets are China's DeepSeek V4.1 Flash and GLM 5.3(max).

Kim I-reun, head of LG AI Research Institute's ExaOne Lab, said, "As a result of internal analysis, we confirmed that the performance of 24 evaluation indicators has improved by more than 70% compared to the model released in August." He added, "Through this technology verification model, we have secured core learning technology to respond to the rapidly changing global benchmark environment."

LG AI Research Institute is also pursuing cooperation with NVIDIA.

We are considering applying NVIDIA's 'Dynamo' technology to increase the inference processing efficiency and service scalability of AI models.

Lim Woo-hyung, co-director of LG AI Research Institute, said, “We will focus on completing the ExaOne model with frontier-level performance by January next year.”