LG AI Research has introduced EXAONE Deep, a groundbreaking AI model designed to push the boundaries of reasoning in mathematics, science, and coding.
EXAONE Deep: A New Era of AI Reasoning
The development of AI models capable of complex reasoning has been a challenge tackled by only a select few global organizations. With EXAONE Deep, LG aims to compete with leading AI models, offering an unprecedented level of reasoning ability.
LG AI Research has significantly enhanced the model’s performance across multiple domains, ensuring it not only excels in specialized fields but also demonstrates a broad understanding of various subjects.
Outstanding Performance Benchmarks
The latest performance benchmarks reveal impressive results:
- Mathematics: The EXAONE Deep 32B model outperformed a significantly larger competitor, despite being only 5% of its size. Meanwhile, the 7.8B and 2.4B versions secured the top spots in all major mathematics benchmarks for their respective tiers.
- Science and Coding: The 7.8B and 2.4B models dominated the rankings in scientific reasoning and coding assessments.
- MMLU (Massive Multitask Language Understanding): The 32B model achieved an impressive 83.0 score, positioning it as the highest-performing domestic Korean model.
International Recognition
Shortly after its debut, EXAONE Deep was acknowledged by Epoch AI, a respected U.S.-based research organization, as one of the most notable AI models. This recognition places EXAONE Deep alongside its predecessor, EXAONE 3.5, marking LG as the only Korean company to have models featured in this prestigious listing for two consecutive years.
Mastery in Mathematics
Across multiple evaluations, LG’s model has demonstrated exceptional mathematical reasoning capabilities:
- The 32B model scored 94.5 in a general mathematics competency test and 90.0 in the American Invitational Mathematics Examination (AIME) 2024.
- In the 2025 AIME, the EXAONE Deep 32B model matched the performance of DeepSeek-R1, a vastly larger 671B model.
- The 7.8B model secured 94.8 on the MATH-500 benchmark and 59.6 in AIME 2025, while the 2.4B model achieved 92.3 and 47.9, respectively.
Advancing Science and Coding
EXAONE Deep has also demonstrated remarkable proficiency in professional science reasoning and software development:
- The 32B model scored 66.1 in the GPQA Diamond test, which assesses problem-solving capabilities in doctoral-level physics, chemistry, and biology.
- Within LiveCodeBench, a coding proficiency benchmark, it achieved a score of 59.5, proving its potential for high-level applications.
- The 7.8B and 2.4B models also emerged as top performers in their categories for both GPQA Diamond and LiveCodeBench.
Enhanced General Knowledge
Beyond its specialized strengths, EXAONE Deep has also showcased an improved understanding of general knowledge, scoring an impressive 83.0 on the MMLU benchmark—solidifying its position as a leading domestic AI model.
The Future of AI Reasoning
LG AI Research envisions EXAONE Deep as a transformative model that will drive AI innovation forward, supporting increasingly complex problem-solving tasks and expanding AI’s role in simplifying and enhancing human life.
For a deeper look into AI’s growing role in security and development, check out how Logicalis US has been recognized for its AI-driven security advancements.
With EXAONE Deep, LG AI Research is setting new standards in AI reasoning, paving the way for future breakthroughs in AI-driven education, research, and enterprise applications.
