로그인|회원가입|고객센터|기업교육 문의
페이지 맨 위로 이동
검색

AI

LG Releases In-House AI Model Exaone 2.5 for Free

Dong-A Ilbo | Updated 2026.11.10
70% performance improvement over the model from 8 weeks ago
Internal evaluation shows it is on par with DeepSeek
LG AI Research Institute fully opened the K-Exaone 2.5 API free of charge on the 9th. (Provided by LG)
On the 9th, LG AI Research Institute made it possible for anyone to connect its artificial intelligence (AI) model “K-Exaone 2.5” to services they are developing. Once an authentication key is issued via the institute’s website, the model can be used free of charge through an application programming interface (API).

The institute unveiled K-Exaone 2.5 on the same day at “NVIDIA AI Day Seoul 2026,” held at COEX in Gangnam-gu, Seoul. This model was created by enhancing the performance of the second model released in August under the institute’s independent AI foundation model development project (“Dokpamo”). Developers and companies can connect this improved model to services under development to verify how well it handles actual business tasks.

According to an internal evaluation based on 24 indicators, the overall performance of the new model improved by more than 70% compared with the second model disclosed eight weeks earlier.

The model also showed comparable performance against Chinese models. It was benchmarked against DeepSeek’s “DeepSeek V4.1 Flash” and Zhipu AI (Z.ai)’s “GLM 5.3 Max” by dividing into four categories the AI’s capability to autonomously plan and use tools to complete tasks.

In “Automation Bench,” which evaluates the ability to use productivity software, K-Exaone 2.5 scored 74.6 points, slightly trailing DeepSeek (74.9 points) but ahead of GLM (71.6 points. In “BrowseComp” (78.0 points), which assesses web information retrieval, as well as problem solving during tasks (77.3 points) and software development and modification (59.2 points), it outperformed DeepSeek but trailed GLM.

The institute is also reviewing the adoption of NVIDIA’s “Dynamo,” which distributes user requests across multiple semiconductors. Lim Woo-hyung, co-head of LG AI Research Institute, said, “The institute will focus on completing an Exaone model with frontier-level performance by January next year.”

Kim Jae-hyeong

AI-translated with ChatGPT. Provided as is; original Korean text prevails.
Popular News

경영·경제 질문은 AI 비서에게,
무엇이든 물어보세요.

Click!