EXAONE 3.0
Open bilingual language model from LG AI Research
LG AI Research released EXAONE 3.0 in August 2024 as its first open large language model. The 7.8 billion parameter instruction-tuned version supports English and Korean, aiming to aid open research while delivering strong performance in general tasks and complex reasoning.
- First open model from LG AI Research
- Contains 7.8 billion parameters
- Released in August 2024
- Excels in Korean language tasks
- Instruction-tuned for general use
What it is
EXAONE 3.0 is a large language model developed by LG AI Research. It represents the first time the organisation has made one of its models openly available to the public. The specific version released for download contains 7.8 billion parameters and has been tuned to follow instructions. This release aims to support open research and innovation in the field of artificial intelligence. The model is designed to handle general tasks and complex reasoning problems effectively.
How it came about
The model was introduced in August 2024 through a technical paper. LG AI Research chose to publicly release the 7.8 billion parameter instruction-tuned variant. This move marked a shift towards openness for the EXAONE family. The developers stated their hope that the model would contribute to advancements in expert artificial intelligence systems. The paper was revised in January 2026, but the initial release date remains August 2024.
Inside the model
This is an instruction-tuned language model, meaning it has been trained to respond to specific user commands rather than just predicting the next word. It operates as a decoder-only architecture typical of modern large language models. The training process focused on enhancing its ability to follow instructions accurately. While specific details on layers or attention mechanisms are not provided in the summary, the model is optimised for real-world effectiveness. It supports bilingual processing, with a particular emphasis on Korean language capabilities alongside English.
What it can do
Evaluations across various public and internal benchmarks show competitive performance. The model excels particularly in Korean language tasks compared to other open models of similar size. It also achieves compelling results in general tasks and complex reasoning scenarios. The developers claim it holds its own against other state-of-the-art open models with comparable parameter counts. Its strength lies in bilingual proficiency and practical application in real-world settings.
How to use it
The 7.8 billion parameter instruction-tuned model is openly available. LG AI Research released the weights to promote open research. Users can access the model through the provided URL in the technical paper. This open access allows researchers and developers to study, modify, and build upon the model without restrictive barriers. It stands as the first open model in the EXAONE family.
Questions people ask about EXAONE 3.0
Who developed EXAONE 3.0?
LG AI Research developed EXAONE 3.0. It is the first open model released by this organisation, marking a significant step in their approach to sharing artificial intelligence technology with the broader research community.
What languages does EXAONE 3.0 support?
The model is bilingual, supporting both English and Korean. It demonstrates particularly strong performance in Korean language tasks, making it a valuable tool for users and researchers working with these two languages.
Is EXAONE 3.0 free to use?
Yes, the 7.8 billion parameter instruction-tuned version is openly available. LG AI Research released the weights to promote open research and innovation, allowing users to download and utilise the model freely.
Good reading and sources
There is no English Wikipedia article on EXAONE 3.0 yet, so the facts on this page come from the sources listed above: arxiv.org/abs/2408.03541. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.