LG AI Research said on the 9th at Nvidia AI Day Seoul 2026 at COEX in Seoul that it will open the application programming interface (API) for free so anyone can use K-EXAONE 2.5, its upgraded in-house artificial intelligence (AI) foundation model.
LG AI Research Institute says on the 9th it opens the K-EXAONE 2.5 API completely free. /Courtesy of LG
K-EXAONE 2.5 is a technology validation model that applies additional training to the second iteration of the in-house foundation model unveiled in August. LG AI Research said it also confirmed the potential for performance gains by advancing the training methodology of the existing model ahead of developing a new third iteration.
Developers, startups, companies, and research institutes can use K-EXAONE 2.5 for service development and research at no cost by simply getting an API key. LG AI Research said it decided to open the API for free to give users a chance to try a Korean AI model directly and apply it to their services.
LG AI Research also said it is considering applying Nvidia’s Dynamo technology to improve inference efficiency and service scalability for its AI model.
LG AI Research also released results for four performance metrics that evaluate agentic AI capabilities, in which AI plans on its own, uses tools, and completes real tasks. It compared the model with DeepSeek V4.1 Flash and GLM 5.3 (max), China’s representative frontier-level open-weight models.
K-EXAONE 2.5 scored 74.6 on AutomationBench, which evaluates the ability to automate tasks using software-as-a-service (SaaS) tools, beating GLM at 71.6 and performing at a level similar to DeepSeek (74.9).
On ClioEval, which measures an AI agent’s problem-solving ability in work settings, it scored 77.3 (DeepSeek 75.8, GLM 77.6); on BrowseComp, which assesses web navigation to find needed information, it scored 78.0 (DeepSeek 75.8, GLM 79.0); and on SWE-Bench Pro, which evaluates the ability to develop and modify software, it scored 59.2 (DeepSeek 58.4, GLM 64.0).
Kim I-reun, head of the EXAONE Lab at LG AI Research, said, “Our internal analysis shows that overall performance across 24 evaluation metrics improved by more than 70% compared with the model released in August.”
※ This article has been translated by AI. Share your feedback here.