EXAONE 3.5
LG AI Research open language model series for real-world tasks
LG AI Research released EXAONE 3.5 in December 2024 as a series of instruction-tuned language models. Available in three sizes, the models focus on long-context understanding and practical instruction following for research and commercial use.
- Released by LG AI Research in December 2024
- Available in 32B, 7.8B, and 2.4B sizes
- Top scores in seven instruction-following benchmarks
- Leading performance in four long-context tests
- Open for research, commercial licence via contact
What it is
EXAONE 3.5 is a family of large language models created by LG AI Research. The series launched in December 2024 with three distinct sizes: 32 billion, 7.8 billion, and 2.4 billion parameters. These models are designed to handle real-world scenarios through strong instruction following. They support long-context comprehension and perform competitively against other open models of similar scale. The project aims to provide robust tools for both academic study and industrial application.
How it came about
LG AI Research developed the EXAONE 3.5 series to address practical use cases in language processing. The team published a technical report in December 2024 detailing the model capabilities. This release followed earlier work in the EXAONE line, focusing on improving performance in complex tasks. The developers updated the documentation in January 2026 to reflect ongoing refinements. The project represents a continued effort by the South Korean lab to contribute open resources to the global AI community.
Inside the model
The models use an instruction-tuned architecture optimised for following user directions. LG AI Research trained the systems to excel in long-context understanding, allowing them to process extensive documents. The design prioritises performance across general benchmarks while maintaining efficiency in smaller configurations. The 32 billion parameter version serves as the largest option, while the 2.4 billion variant offers a lightweight alternative. Training focused on achieving high scores in specific evaluation areas such as instruction adherence and context retention.
What it can do
EXAONE 3.5 demonstrates strong performance in instruction following, topping seven benchmarks in this category. It also leads in four benchmarks dedicated to long-context comprehension. The models show competitive results across nine general benchmarks when compared to similar open-source alternatives. These metrics suggest the system handles diverse linguistic tasks effectively. The varying sizes allow users to balance computational cost with performance needs for different applications.
How to use it
LG AI Research makes the model weights available for download for research purposes. Users can access the files through public repositories linked in the technical report. For commercial deployment, organisations must contact LG AI Research directly via their official email. This dual approach supports academic exploration while managing enterprise licensing separately. The open availability encourages broader testing and adaptation by the developer community.
Questions people ask about EXAONE 3.5
Who developed EXAONE 3.5?
LG AI Research developed and released the EXAONE 3.5 series. The team published a technical report describing the models and their performance metrics in late 2024.
What sizes are available for EXAONE 3.5?
The series includes three configurations with 32 billion, 7.8 billion, and 2.4 billion parameters. This range allows users to select a model based on their hardware constraints.
Can I use EXAONE 3.5 for commercial products?
Commercial use requires direct contact with LG AI Research. The models are freely available for research purposes, but businesses must negotiate terms through the official contact channel.
Good reading and sources
There is no English Wikipedia article on EXAONE 3.5 yet, so the facts on this page come from the sources listed above: arxiv.org/abs/2412.04862. 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.