top of page

LG Unveils Korea's First Open-weight Hybrid AI, 'EXAONE 4.0'

Proving Superior Performance in Benchmarks Against Leading Open-Weight Models from the US, China, and France.


ree

On July 15, the LG AI Research unveiled EXAONE 4.0, the first open-weight hybrid AI model in Korea. EXAONE 4.0 is a groundbreaking model that integrates a Large Language Model (LLM), which excels in natural language processing, generation, and knowledge-based rapid responses, with a reasoning AI model capable of forming and verifying hypotheses to solve problems.


Currently, only a few global players, such as Anthropic (developer of Claude) in the US and Alibaba (developer of Qwen) in China, have released hybrid AI models. OpenAI has also stated that it is developing GPT-5 as a hybrid AI, an integrated model. Following the release of EXAONE Deep, Korea's first reasoning AI model, in March, LG AI Research has now unveiled EXAONE 4.0, the country's first hybrid AI model, just four months later, further enhancing its competitiveness in the global AI market.


In particular, 'EXAONE 4.0' has proven itself to be world-class, surpassing leading open-weight models from the United States, China, and France in benchmark comparisons evaluating AI model performance. EXAONE 4.0 achieved scores of 92.3/81.8 on MMLU-Redux/MMLU-Pro (World knowledge), 66.7 on LiveCodeBench v6 (Code), 75.4 on GPQA-Diamond (Science), and 85.3 on AIME 2025 (Mathematics).


LG AI Research has released EXAONE 4.0 as an open-weight model on Hugging Face, a global open-source AI platform, for research, academic, and educational purposes. While open-weight models do not disclose the AI model's design or training data, they do reveal the weights, which allow for understanding the AI's data processing methods, enabling modification or redistribution. Examples of prominent open-weight models include Google's Gemma, Meta's Llama, Microsoft's Phi from the US, Alibaba's Qwen from China, and Mistral AI's Mistral from France.

EXAONE 4.0 (32B) benchmarked against open-weight models
EXAONE 4.0 (32B) benchmarked against open-weight models

"LG AI Research will continue our research and development to ensure EXAONE becomes a leading frontier model representing Korea, proving its competitiveness in the global market," said Jinsik Lee, Head of EXAONE Lab at LG AI Research. LG AI Research plans to hold the 'LG AI Talk Concert 2025' on July 22 at LG Sciencepark in Magok, Seoul, where it will announce its AI technology research and development achievements, including EXAONE 4.0, and future plans.


LG AI Research has unveiled a 32B (32 billion parameters) expert model and a 1.2B (1.2 billion parameters) on-device model. The 32B expert model demonstrated a high level of expertise by passing the written exams for six national professional qualifications, including doctor, dentist, herbal pharmacist, customs broker, appraiser, and Insurance adjuster. LG AI Research expects EXAONE 4.0 to provide highly accurate answers in fields requiring specialized knowledge, delivering essential and exceptionally valuable solutions to customers.


LG AI Research has focused on developing a practical on-device model that can be directly utilized in various devices such as home appliances, smartphones, automotive infotainment systems, and robots. On-device models offer strong advantages in terms of privacy protection and security, as they can process information quickly and securely within the electronic device without connecting to external servers. This model is half the size of the EXAONE 3.5 2.4B model released last December, making it lighter and more economical. Despite its smaller size, it outperformed OpenAI's GPT-4o mini in specialized evaluation metrics for mathematics, coding, and science, demonstrating the highest level of performance among AI models of similar scale worldwide. LG AI Research plans to take the lead in the rapidly expanding on-device AI market with EXAONE.


To popularize AI technology, LG AI Research has partnered with FriendliAI, Hugging Face's official AI model distribution partner, to launch the API service for EXAONE 4.0. This partnership allows anyone, from individual developers to enterprises, to easily utilize EXAONE or integrate it into their services without the need for high-performance GPUs. LG AI Research expects the API release to be a significant turning point, accelerating the adoption of AI technology across various industries.


On July 15, LG AI Research hosted 'EXAONE Partner's Day', where it discussed ways to activate cooperation within the EXAONE ecosystem with 22 partner companies. Meanwhile, LG Corporation Chairman and CEO Kwang-mo Koo, emphasized in his New Year's address this year, "The many moments of challenging new areas and creating unprecedented value have built today's LG." He added, "We will make advanced technologies like AI conveniently usable in daily life, creating a new lifestyle where valuable time can be spent on more enjoyable and meaningful activities."

Comments


connexion_panel_edited.jpg
CXO_8-in-1.png
subscribe_button.png

 

Disclaimer:

The information contained in this site is for reference only. While we have made every attempt to ensure that the information contained in this site has been obtained from reliable sources, we are not responsible for any errors or omissions, or for the results obtained from the use of this information. All information in this site is provided "as is", with no guarantee of completeness, accuracy, timeliness or of the results obtained from the use of this information, and without warranty of any kind, express or implied, including, but not limited to warranties of performance, merchantability and fitness for a particular purpose. In no event will Ho Hon Asia Limited, its related partnerships or corporations, or the partners, agents or employees thereof be liable to you or anyone else for any decision made or action taken in reliance on the information in this site or for any consequential, special or similar damages, even if advised of the possibility of such damages.
Certain links in this site connect to other websites maintained by third parties over whom we have no control. We make no representations as to the accuracy or any other aspect of information contained in other websites.

2025 @ Inno-Thought and its affiliates. All rights reserved.

bottom of page