How can developers and companies access LG AI Research’s K-EXAONE 2.5 API, and what does it cost?

Developers, startups, companies and research institutes can obtain an API key through LG AI Research’s platform, docs.exaone.ai. The model is available at no cost for use in services and research.

What is K-EXAONE 2.5, and how does it relate to Korea’s sovereign AI project?

K-EXAONE 2.5 is a technology verification model created by applying additional training to K-EXAONE 2.0. The earlier model was the second one LG AI Research developed under a government project intended to build homegrown AI foundation models and reduce Korea’s reliance on foreign systems.

LG AI Research opens K-EXAONE 2.5 API free of charge

Q: How did K-EXAONE 2.5 perform against DeepSeek V4.1 Flash and GLM 5.3 on AutomationBench?

In LG AI Research’s tests on open datasets, K-EXAONE 2.5 scored 74.6 on AutomationBench, which measures task automation using software-as-a-service tools. DeepSeek V4.1 Flash scored 74.9 and GLM 5.3 (max) scored 71.6.

Listen

Summary

LG AI Research opened the K-EXAONE 2.5 API free of charge at NVIDIA AI Day Seoul 2026 at Coex in southern Seoul on Monday. Developers, startups, companies and research institutes can use the model in services and research by obtaining an API key through the company’s platform. The model is a technology verification version built by further training K-EXAONE 2.0. LG AI Research plans to apply validated training techniques to its third-round model and aims to complete an EXAONE model with frontier-level performance by January next year.

Key Facts

LG AI Research said its internal analysis found performance across 24 benchmarks improved by more than 70 percent compared with the model released in July.
On AutomationBench, K-EXAONE 2.5 scored 74.6, compared with 74.9 for DeepSeek V4.1 Flash and 71.6 for GLM 5.3 (max).
K-EXAONE 2.5 scored 77.3 on Claw-Eval, 78.0 on BrowseComp and 59.2 on SWE-Bench Pro.
LG AI Research is reviewing the use of NVIDIA’s Dynamo technology to improve inference efficiency.

By Jhoo Dong-chan

Published Nov 9, 2026 2:29 pm KST

A promotional image of LG AI Research's K-EXAONE 2.5 platform / Courtesy of LG AI Research

LG AI Research, the artificial intelligence (AI) research arm of LG Group, opened the application programming interface (API) for its upgraded EXAONE 2.5 foundation model free of charge, reporting a performance boost of over 70 percent in just eight weeks.

The company announced the move at NVIDIA AI Day Seoul 2026, held at Coex in southern Seoul on Monday. Starting the same day, developers, startups, companies and research institutes can use the model in their services and research at no cost by obtaining an API key through the company’s platform, docs.exaone.ai.

K-EXAONE 2.5 is a technology verification model built by applying additional training to K-EXAONE 2.0, the second model LG AI Research developed under the government’s sovereign AI foundation model project, which aims to build homegrown AI models that reduce Korea’s reliance on foreign systems.

According to the company’s own tests on open datasets, K-EXAONE 2.5 performed on par with leading Chinese open-weight models DeepSeek V4.1 Flash and GLM 5.3 (max) in agentic AI, in which AI plans tasks and uses tools on its own to complete real work.

It scored 74.6 on AutomationBench, which measures task automation using software-as-a-service tools, compared with 74.9 for DeepSeek and 71.6 for GLM. On Claw-Eval, BrowseComp and SWE-Bench Pro, which test workplace problem-solving, web browsing and software development, it scored 77.3, 78.0 and 59.2, respectively.

“Internal analysis showed that overall performance across 24 benchmarks improved by more than 70 percent compared with the model released in July,” said Kim Yi-reun, head of the EXAONE Lab.

The company is also reviewing the use of NVIDIA’s Dynamo technology to improve inference efficiency.

LG AI Research plans to apply the training techniques validated through K-EXAONE 2.5 to its third-round model.

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

This article was published with the assistance of generative AI and edited by The Korea Times.

Explore More