LG AI Research has introduced two AI foundation models designed for manufacturing, including production-data analysis and automated visual inspection.

The models, EXAONE Tabular and EXAONE Omni-Inspect, were presented at LG AI Talk Concert 2026 in Seoul. LG said they are designed to work with changing production conditions while reducing the need for additional model training.

The announcement also builds on LG AI Research’s existing work on enterprise AI infrastructure. Its EXAONE portfolio includes an on-premise system designed to keep AI models and enterprise data within company-controlled environments, although the recent announcement does not link that system directly to Tabular or Omni-Inspect.

AI models built for changing factory conditions

EXAONE Tabular is designed to analyse structured data, including information about manufacturing processes and product quality. It examines relationships within numerical data and uses them to make predictions about production conditions.

LG said the model can make predictions using relatively small amounts of data when manufacturers introduce new production environments or change existing processes. The company said this can reduce the data collection and retraining required when production conditions change.

EXAONE Tabular uses in-context learning rather than training a separate version of the underlying model for every dataset. LG’s model documentation says labelled examples can be provided before the model predicts new data without dataset-specific gradient updates or additional training.

LG said this approach allows the model to work with small amounts of new data when production conditions change. The Korea Times reported that the company reduced the time needed to respond to manufacturing model changes by 85%.

LG has previously described EXAONE Tabular as a tabular foundation model trained on more than one billion synthetic tabular data examples. The company said the model can also estimate missing values in industrial datasets by analysing surrounding information.

The model is also smaller than several foundation models used for comparable tabular-data tasks. LG’s released documentation lists 20.8 million parameters for its classification version and 21.1 million for regression.

In an LG AI Research technical report, the company compared its regression model with Google’s 1.64-billion-parameter TabFM and said EXAONE Tabular reached a similar performance range at about one-eleventh of the inference cost. The comparison is based on LG’s own benchmark results.

LG’s published TabArena results put median prediction time at 0.605 seconds per 1,000 samples, compared with 6.985 seconds for TabFM under the listed benchmark configurations. LG recommends a CUDA-capable GPU for EXAONE Tabular, although CPU inference is also supported at a lower speed.

EXAONE Omni-Inspect addresses a separate manufacturing workload: visual quality inspection. The system analyses camera images to identify defects in components and products during manufacturing.

LG said the model can continue performing inspections when the appearance of a product or manufacturing process changes without requiring the entire system to be retrained. This reduces the need to update inspection models each time factories introduce new products or modify production processes, according to the company.

Tabular works with structured production and quality data, while Omni-Inspect handles visual inspection data. The Korea Times reported that Omni-Inspect is designed to continue identifying defects even when new products or manufacturing processes change the images it receives.

The company is also developing a vision inspection agent that handles data sampling, labelling, and model training. LG said the system is being developed to automate more of the process required to adapt machine-vision inspection systems to changing manufacturing conditions.

LG AI Research said it has worked on more than 100 industrial problems since it was established in December 2020. These include battery life and capacity prediction, defective-product detection, and production and materials planning.

“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,” LG AI Research co-head Lim Woo-hyung said. Lim said industrial systems need to account for multiple variables and exceptional cases that general-purpose models are not necessarily designed to handle.

Bringing industrial AI closer to production data

LG introduced EXAONE On-Premise in July 2025 as a full-stack system that can run within a company’s own infrastructure. The company said it was designed for organisations that need to use AI while keeping sensitive information inside their own environments.

Yonhap described EXAONE On-Premise as a full-stack solution that allows companies to build secure, in-house agentic AI systems. IEEE Spectrum reported that LG’s longer-term goal is to enable enterprises to run autonomous agents securely within their own infrastructure.

LG’s recent announcement does not specify whether Tabular or Omni-Inspect will run through EXAONE On-Premise or on factory-floor edge infrastructure.

Edge architectures already process some industrial data closer to the equipment where it is generated. The US National Institute of Standards and Technology describes the industrial “intelligent edge” as combining computing, analytics, and connectivity closer to where data is captured.

NIST says data at the edge can be captured, encrypted, integrated, processed, and stored close to where it is generated. It describes a hybrid architecture in which AI can respond to process inefficiencies and quality defects at the edge while ongoing learning takes place at the core.

LG’s manufacturing work also extends into robotics. The company is developing a robot foundation model intended to support automated factory systems by enabling robots to interpret conditions, make decisions, and carry out physical actions.

LG said the research includes algorithms intended to prevent accidents during robot operation, with safety included as part of its work on robot foundation models.

LG ultimately plans to use robot foundation models as part of autonomous factory systems that coordinate equipment across production environments. Its stated goal is to move beyond automating individual robots toward systems that can coordinate operations across an entire plant.

(Photo by Homa Appliances)

See also: Skild trains S1 robot physical AI model on NVIDIA infrastructure

Want to learn more about the IoT from industry leaders? Check out IoT Tech Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including AI & Big Data Expo and the Cyber Security Expo. Click here for more information.

IoT News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.