LG AI Research: “AI performance up 70% in eight weeks” ... Opens K-EXAONE 2.5 for free

[Financial News] LG AI Research has set out to narrow the technological gap with global frontier-level AI models, improving the performance of its proprietary artificial intelligence (AI) foundation model by at least 70% in just eight weeks. It plans to make the improved K-EXAONE 2.5 freely available to test its practical usability and, based on the results, complete development of a next-generation model by next January.
LG AI Research said on the 9th at NVIDIA AI Day Seoul 2026, held at COEX in Seoul, that it would make the K-EXAONE 2.5 application programming interface (API) available for free.
K-EXAONE 2.5 is a technical validation model created by applying additional training to the second model in LG AI Research’s proprietary AI foundation model project, unveiled last August. LG AI Research improved its performance by refining the training methodology used for the existing model.
According to LG AI Research, K-EXAONE 2.5 achieved performance competitive with global AI models such as China’s DeepSeek V4.1 Flash and GLM 5.3 (max) in its own evaluation using public datasets.
In an assessment of agentic AI capabilities—the ability of AI to make plans on its own and use tools to perform tasks—K-EXAONE 2.5 scored 74.6 on AutomationBench, which measures work-automation ability, a result comparable to DeepSeek’s 74.9. On Claw-Eval, which assesses problem-solving ability in work environments, it scored 77.3, surpassing DeepSeek’s 75.8.
K-EXAONE 2.5 scored 78.0 on BrowseComp, which evaluates web-browsing ability, and 59.2 on SWE-Bench Pro, which measures software development ability. Both scores were higher than DeepSeek’s.
Kim I-reun, head of Exaone Lab at LG AI Research, said, “Our internal analysis confirmed that performance aggregated across 24 evaluation metrics had improved by at least 70% compared with the model released in August,” adding, “We have laid the groundwork for achieving frontier-level performance.”
Starting today, LG AI Research is making the K-EXAONE 2.5 API available for free through its own platform. Developers, startups, companies, research institutions and others can obtain an API key and use it for service development and research at no additional cost.
It is also pursuing technical cooperation with NVIDIA. LG AI Research is considering applying NVIDIA’s Dynamo technology to improve the efficiency of AI model inference and the scalability of its services.
LG AI Research plans to incorporate into development of its third model the proprietary training techniques validated with this model, including improvements to data quality, post-training and reinforcement learning. Its goal is to complete an AI foundation model with global frontier-level performance by next January.
Lim Woo-hyung, co-director of LG AI Research, said, “We are currently focusing our efforts on developing the third model, and we hope that with this API release, more developers and companies will experience the improved performance firsthand,” adding, “We will focus on completing an EXAONE model with frontier-level performance by next January.” 
[email protected] Yoon Hong-jip Reporter