LG EXAONE Deep is a maths, science, and coding buff

LG AI Analysis has unveiled EXAONE Deep, a reasoning type that excels in advanced problem-solving throughout maths, science, and coding.

The corporate highlighted the worldwide problem in developing complicated reasoning fashions, noting that recently, just a handful of organisations with foundational fashions are actively pursuing this advanced house. EXAONE Deep goals to compete immediately with those main fashions, showcasing a aggressive point of reasoning skill.

LG AI Analysis has targeted its efforts on dramatically bettering EXAONE Deep’s reasoning functions in core domain names. The type additionally demonstrates a robust skill to grasp and follow wisdom throughout a broader vary of topics.

The efficiency benchmarks launched by way of LG AI Analysis are spectacular:

• Maths: The EXAONE Deep 32B type outperformed a competing type, regardless of being handiest 5% of its dimension, in a tough arithmetic benchmark. Moreover, the 7.8B and a pair of.4B variations accomplished first position in all main arithmetic benchmarks for his or her respective type sizes.
• Science and coding: In those spaces, the EXAONE Deep fashions (7.8B and a pair of.4B) have secured the tip spot throughout all main benchmarks.
• MMLU (Large Multitask Language Working out): The 32B type accomplished a rating of 83.0 at the MMLU benchmark, which LG AI Analysis claims is the most efficient efficiency amongst home Korean fashions.

The functions of the EXAONE Deep 32B type have already garnered world reputation.

In a while after its unencumber, it used to be incorporated within the ‘Notable AI Fashions’ record by way of US-based non-profit analysis organisation Epoch AI. This list puts EXAONE Deep along its predecessor, EXAONE 3.5, making LG the one Korean entity with fashions featured in this prestigious record up to now two years.

Maths prowess

EXAONE Deep has demonstrated outstanding mathematical reasoning abilities throughout its more than a few type sizes (32B, 7.8B, and a pair of.4B). In tests according to the 2025 instructional 12 months’s arithmetic curriculum, all 3 fashions outperformed world reasoning fashions of similar dimension.

The 32B type accomplished a rating of 94.5 in a basic arithmetic competency take a look at and 90.0 within the American Invitational Arithmetic Exam (AIME) 2024, a qualifying examination for the United States Mathematical Olympiad.

Within the AIME 2025, the 32B type matched the efficiency of DeepSeek-R1—a considerably greater 671B type. This outcome showcases EXAONE Deep’s environment friendly finding out and robust logical reasoning talents, specifically when tackling difficult mathematical issues.

The smaller 7.8B and a pair of.4B fashions additionally accomplished height scores in main benchmarks for light-weight and on-device fashions, respectively. The 7.8B type scored 94.8 at the MATH-500 benchmark and 59.6 on AIME 2025, whilst the two.4B type accomplished rankings of 92.3 and 47.9 in the similar opinions.

Science and coding excellence

EXAONE Deep has additionally showcased outstanding functions in skilled science reasoning and device coding.

The 32B type scored 66.1 at the GPQA Diamond take a look at, which assesses problem-solving abilities in doctoral-level physics, chemistry, and biology. Within the LiveCodeBench analysis, which measures coding skillability, the type accomplished a rating of 59.5, indicating its doable for high-level packages in those skilled domain names.

The 7.8B and a pair of.4B fashions persevered this pattern of robust efficiency, each securing first position within the GPQA Diamond and LiveCodeBench benchmarks inside of their respective dimension classes. This fulfillment builds upon the good fortune of the EXAONE 3.5 2.4B type, which in the past crowned Hugging Face’s LLM Readerboard within the edge department.

Enhanced basic wisdom

Past its specialized reasoning functions, EXAONE Deep has additionally demonstrated progressed efficiency typically wisdom working out.

The 32B type accomplished an outstanding rating of 83.0 at the MMLU benchmark, positioning it because the top-performing home type on this complete analysis. This means that EXAONE Deep’s reasoning improvements lengthen past explicit domain names and give a contribution to a broader working out of more than a few topics.

LG AI Analysis believes that EXAONE Deep’s reasoning developments constitute a bounce in opposition to a long run the place AI can take on more and more advanced issues and give a contribution to enriching and simplifying human lives via steady analysis and innovation.

See additionally: Baidu undercuts rival AI fashions with ERNIE 4.5 and ERNIE X1

Wish to be told extra about AI and large knowledge from trade leaders? Take a look at AI & Giant Knowledge Expo going down in Amsterdam, California, and London. The great tournament is co-located with different main occasions together with Clever Automation Convention, BlockX, Virtual Transformation Week, and Cyber Safety & Cloud Expo.

Discover different upcoming endeavor generation occasions and webinars powered by way of TechForge right here.