# LG Unveils Two AI Foundation Models Tailored for Smart Manufacturing
**Published Date: February 2026**
LG AI Research has officially launched two specialized artificial intelligence foundation models aimed at transforming how manufacturers handle data analysis and quality control on the production floor. The models, named EXAONE Tabular and EXAONE Omni-Inspect, were debuted at LG AI Talk Concert 2026 in Seoul, signaling the company’s deepening commitment to enterprise-grade AI solutions for industrial environments.
—
## Designed to Adapt, Not Just Analyze
What sets these two models apart is their emphasis on adaptability. Rather than requiring constant retraining every time a factory modifies its workflows or introduces new products, both models are engineered to function effectively under shifting production conditions with minimal additional effort.
EXAONE Tabular focuses on structured data — the kind of numerical, tabular information that underpins decisions about manufacturing processes and product quality. It identifies relationships within datasets and generates predictions based on those patterns. A notable strength is its ability to deliver meaningful results even when manufacturers have only limited historical data available for a new production environment.
The model achieves this through a technique known as in-context learning. Instead of retraining the entire model for each new dataset, users can supply labeled examples directly in the input, and the model uses those examples to inform its predictions. No dataset-specific gradient updates or additional training cycles are necessary.
## A Lightweight Model with Competitive Performance
From a technical standpoint, EXAONE Tabular is remarkably compact. The classification variant contains 20.8 million parameters, while the regression version has 21.1 million. This is a fraction of the size of many comparable foundation models in the tabular data space.
In LG’s own benchmarks, the regression model was pitted against Google’s TabFM, which boasts 1.64 billion parameters. According to LG’s internal testing, EXAONE Tabular achieved a similar range of predictive accuracy but at roughly one-eleventh of the inference cost. Per 1,000 samples, median prediction time was recorded at 0.605 seconds, compared to 6.985 seconds for TabFM under identical benchmark settings.
While LG recommends running the model on a CUDA-capable GPU for optimal speed, CPU-based inference remains supported, albeit at a slower pace. This flexibility gives manufacturers options when integrating the model into existing hardware setups.
## Visual Inspection Gets an AI Upgrade
EXAONE Omni-Inspect tackles a different but equally critical manufacturing challenge: automated visual quality inspection. Using camera imagery, the model identifies defects in components and finished products as they move through the production line.
The key advantage here is continuity. When a factory changes products or adjusts its manufacturing process, the visual characteristics of items on the line also change. Most inspection systems require a full retraining cycle in such scenarios. LG claims Omni-Inspect can continue identifying defects effectively under these new conditions without a complete overhaul of the underlying model.
Additionally, LG AI Research is developing a vision inspection agent — a system designed to automate several stages of the inspection pipeline, including data sampling, labeling, and model training. The goal is to reduce the human intervention typically required to keep machine-vision systems aligned with evolving production realities.
## A Broader Enterprise AI Vision
The introduction of these two models is part of a larger strategy. LG AI Research has been building out the EXAONE portfolio since its establishment in December 2020, and the company has applied its AI capabilities to more than 100 industrial problems to date. These include battery life and capacity prediction, defective product detection, and production and materials planning.
Earlier in 2025, the company introduced EXAONE On-Premise, a full-stack system that allows businesses to run AI models entirely within their own infrastructure. This approach keeps sensitive enterprise data inside company-controlled environments, a priority for many manufacturers dealing with proprietary processes and intellectual property.
While the recent model announcements do not explicitly tie Tabular and Omni-Inspect to the EXAONE On-Premise system, the strategic overlap is clear. LG is progressively building an integrated ecosystem of AI tools that can operate securely and efficiently within industrial settings.
## Beyond the Factory Floor: Robotics and Autonomous Systems
LG’s manufacturing ambitions extend beyond data analysis and visual inspection. The company is also working on a robot foundation model designed to help automated factory systems interpret conditions, make decisions, and execute physical tasks. Safety is a central consideration in this work, with LG incorporating algorithms aimed at preventing accidents during robot operation.
The long-term vision is ambitious: LG aims to move beyond automating individual robots and toward coordinating entire production plants through autonomous systems. This would represent a significant leap from single-task automation to holistic, plant-wide intelligent manufacturing.
LG AI Research co-head Lim Woo-hyung emphasized the company’s core philosophy: “Building a good AI model is important, but LG AI Research’s mission is to solve difficult problems that industries have struggled with for years.” He noted that industrial environments involve countless variables and exceptional cases that general-purpose models are not always built to handle.
## Frequently Asked Questions (FAQ)
**What is EXAONE Tabular used for?**
EXAONE Tabular is designed to analyze structured manufacturing data, including production process metrics and product quality indicators. It makes predictions about production conditions by examining relationships within numerical datasets.
**How does EXAONE Tabular handle new production environments?**
It uses in-context learning, which allows it to make predictions based on small amounts of new labeled data without requiring dataset-specific retraining or gradient updates.
**What is EXAONE Omni-Inspect?**
EXAONE Omni-Inspect is a visual inspection model that analyzes camera images to detect defects in components and products during manufacturing. It is designed to maintain accuracy even when products or processes change.
**How small is EXAONE Tabular compared to other models?**
The classification version has 20.8 million parameters, and the regression version has 21.1 million. This is significantly smaller than competing models such as Google’s TabFM, which has 1.64 billion parameters.
**Can EXAONE Tabular run on standard hardware?**
Yes. While LG recommends a CUDA-capable GPU for optimal performance, the model also supports CPU-based inference, making it accessible for a range of industrial setups.
**What is EXAONE On-Premise?**
EXAONE On-Premise is a full-stack AI system introduced in July 2025 that allows enterprises to run AI models and keep their data within company-controlled infrastructure.
**Is there a connection between EXAONE On-Premise and the new models?**
While the announcements do not explicitly link Tabular and Omni-Inspect to EXAONE On-Premise, both fall under the broader EXAONE portfolio and share the goal of secure, enterprise-ready AI deployment.
**What other areas is LG AI Research working on?**
Beyond these two models, LG is developing a robot foundation model for factory automation and aims to build autonomous systems that coordinate operations across entire manufacturing plants.
—
## Conclusion
LG AI Research’s release of EXAONE Tabular and EXAONE Omni-Inspect represents a focused push into practical, industrial AI. By prioritizing adaptability, lightweight architecture, and ease of deployment, these models address real pain points in modern manufacturing — from the cost and time of retraining to the challenge of maintaining quality inspection accuracy across changing production lines. Coupled with the company’s work on on-premise infrastructure and robot foundation models, LG is building toward a vision of intelligent manufacturing that is secure, scalable, and deeply integrated with industrial operations. As factory floors become increasingly data-driven, tools like these are likely to play a central role in shaping the next generation of smart manufacturing.
Thank you for reading
