AI for industrial sites evolves into EXAONE ecosystem

The LG AI Talk Concert 2025 was held recently. At the event, LG AI Research showcased its advanced technology and industrial application capabilities accumulated over the five years since its establishment.

In addition, the EXAONE Ecosystem was unveiled, which encompasses everything from AI models to data, services, and infrastructure based on EXAONE, presenting a concrete vision for the future of AI.

The LG AI Talk Concert 2025, which was the first official event under new leadership, also symbolized an important turning point for LG AI Research. Here’s an account of the exciting events of that day.

LG AI Research's challenges and achievements that connect “first” and “best”
The LG AI Talk Concert 2025 was an opportunity to look back on the journey of LG AI Research with a record of “firsts” and “best” in Korea.

2021: Korea's first independently developed multimodal AI model, EXAONE 1.0, is unveiled

2022: Collaboration between AI and human designers leads to participation in New York Fashion Week and winning a gold medal at an advertising festival

2023: AI Ethics Implementation partnership signs with UNESCO

2024: Korea's first open-source research model, EXAONE 3.0, and EXAONE 3.5, featured in the Stanford AI report, are released

2025: Korea's first inference-based model, EXAONE Deep, and hybrid model, EXAONE 4.0, are announced

At the event, Woohyung Lim, Head of LG AI Research, announced that the EXAONE series has recorded over 5.1 million downloads to date, explaining that this is the highest level of achievement among domestic AI models.

In particular, through “Expert AI for Enterprise: Industrial Value Chain Application Cases,” he provided deep insights into how AI is securing expertise and effectiveness in actual industrial sites.

Based on these achievements, LG AI Research emphasized that it is constantly making strides forward based on three core values of innovation: securing global competitiveness based on its own foundation model, strengthening versatility and expertise through application in various industrial fields, and building an ecosystem in collaboration with partner companies.

New evolution of EXAONE ecosystem
At this year's LG AI Talk Concert 2025, the latest EXAONE AI model and the data and infrastructure supporting it were unveiled, demonstrating that EXAONE is moving beyond a single model to become an integrated solution for solving industry problems.

EXAONE Path 2.0
EXAONE Path 2.0 is a model specialized in histopathological image analysis, which combines multi-omics data containing DNA and RNA information with pathological images for learning.

This enables the rapid prediction of specific genetic mutations using only pathological images, without the need for separate genetic testing, and significantly reduces the diagnosis time, which previously took more than two weeks, to less than one minute, enabling patient-tailored treatment.

In particular, it has innovated by moving beyond the limitations of existing methods that were limited to patch-level analysis, and instead trains on the entire slide, significantly improving accuracy. Currently, EXAONE Path 2.0 is available on the NVIDIA MONAI platform, having recorded over 12,000 downloads and gaining high trust in medical settings.

EXAONE 4.0
Honglak Lee, Head of LG AI Research and CSAI, introduced EXAONE 4.0 as an “innovative hybrid model that combines outstanding inference capabilities and natural language understanding capabilities,” and explained in detail the configuration and performance of the large model with 32B parameters and the lightweight model for on-device use with 1.2B parameters.

Notably, he presented performance evaluation results that surpassed major competing models such as Meta’s Llama, Google’s Gemini, and Anthropic’s Claude in global AI model evaluations, and also mentioned that it has been officially listed on Epoch AI's “Notable AI Models” platform, receiving international recognition.

EXAONE 4.0 VL
The evolution of multimodal AI technology also attracted considerable attention. The EXAONE 4.0 VL, unveiled for the first time at this event, is equipped with advanced visual language capabilities that enable it to accurately interpret not only general images but also tables, charts, and infographics in complex documents.

In particular, it achieved the world's highest performance in ChartQA, a key benchmark for document visual understanding, demonstrating its ability to strongly support data-driven decision-making in companies.

EXAONE service: Intersection of technology and practice
EXAONE's technological capabilities are being realized as core services that can be immediately applied in industrial settings.

ChatEXAONE Beta
ChatEXAONE is a Work Agent service developed to maximize work efficiency within companies. Since its launch in December last year, it has been actively used by more than 50,000 office workers at LG Group. It provides in-depth insights through its inference enhancement mode and in-depth research capabilities, and has received high satisfaction ratings from users.

Now, ChatEXAONE Beta has been launched for general users beyond LG Group employees.

EXAONE Data Foundry
The biggest challenge for companies when adopting AI technology is how to quickly and securely obtain specialized, high-quality data. Utilizing external experts poses significant security risks, such as core data leakage, while relying solely on internal personnel consumes excessive time and costs.

EXAONE Data Foundry is an AI-based data auto-generation platform developed to fundamentally solve these issues, and has demonstrated how it can increase productivity and performance.

EXAONE On-Premise
In line with the growing demand for generative AI that can be integrated with proprietary systems and securely protect corporate data, we introduced the EXAONE On-Premise package, which can be used stably within corporate networks.

This package supports a wide range of infrastructure environments, including GPUs and NPUs, and allows companies to quickly customize configurations by selecting only the features they need, significantly reducing the burden of implementation and operation.

Collaboration with external partners
LG AI Talk Concert 2025 was also a meaningful occasion to strengthen partnerships with external partners. Byeonggon Jeon, CEO of FrendliAI, an official partner of Hugging Face, Joonho Baek, CEO of FuriosaAI, and Arman Sahovic, Head of Asia-Pacific Data Platform Solutions at LSEG, attended the event in person to discuss how AI will change the future.

FriendliAI
Jungkyu Choi, Head of the AI Agent Group, announced a strategic partnership with FriendliAI and the launch of EXAONE 4.0 commercial API services. Byeonggon Jeon, CEO of FriendliAI, demonstrated EXAONE 4.0 in action on stage, and mentioned that FriendliAI's inference infrastructure technically supports the stability and scalability of the service.

FuriosaAI
FuriosaAI has been conducting research to dramatically improve EXAONE’s inference speed based on its proprietary AI semiconductors. Joonho Baek, CEO of FuriosaAI, explained, “Thanks to the combined efforts of both companies, we have significantly improved energy efficiency and processing speed, satisfying the performance requirements necessary to provide EXAONE On-Premise solutions.”

LSEG
LG AI Research has been steadily expanding its open collaboration ecosystem with various global and industry partners. At this event, we introduced our collaboration with LSEG, a global financial information company. Arman Sahovic of LSEG explained why LSEG chose LG AI Research as a collaboration partner and revealed that the two companies are jointly developing a “Master Score” based on advanced prediction technology and actual insights.

Future of AI + NEXUS
n the final session, LG AI Research presented its future vision, “Future of AI + NEXUS.”

NEXUS is a data compliance agent, an Agentic AI designed to resolve copyright disputes over AI model training data, which are increasing due to the rapid development of AI technology. Copyright reviews of training data through NEXUS are 26% more accurate than those conducted by human experts, 45 times faster, and cost 1/1,000th as much. This innovative system is expected to become the centerpiece of safe and responsible future AI technology.

Along with this, Honglak Lee, Head of LG AI Research and CSAI, explained: “AI is evolving beyond simply generating content based on user commands to become an active agent AI that can make its own judgments and take action. Agent AI that understands user goals and establishes action plans will soon advance to become physical AI that directly acts in the real world.”

This signifies that AI is transforming into an entity that moves, measures, and controls in physical spaces through integration with sensors, robotics, and edge computing. LG AI Research also revealed plans to accelerate the development of physical AI technology.

The EXAONE Showroom was unveiled, where visitors can experience LG AI Research's technologies and services firsthand. Many attendees actively explored the showroom and experienced the demos, freely communicating with LG AI Research members and deepening their understanding of EXAONE services.

EXAONE goes beyond a single AI model to expand into a platform, service, and ecosystem, providing substantial AI value across industries. LG AI Research plans to continue to pursue trustworthy AI, AI applicable to industry settings, and AI for professionals, pursuing continuous innovation.