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

LG AI Research unveiled Exaone Deep, a thinking model that excels in solving complex problems through mathematics, science and coding.

The company has highlighted the global challenge in creating advanced thinking models, noting that it is currently, only a few organizations with basic models are followed by the activity of this complex area. Exaone Deep aims to compete directly with these pioneering models, as it displays a competitive level of thinking.

LG AI Research has focused on improving the thinking capabilities in Exaone Deep significantly in the basic fields. The model also shows a strong ability to understand and apply knowledge through a broader set of topics.

LG AI Research performance standards:

• mathematics: The Exaone Deep 32B model outperforms a competitive model, although only 5 % of its size, in the requirement of mathematics. Moreover, 7.8B and 2.4B versions ranked first in all the main mathematics standards for their sizes.
• Science and coding: In these areas, the deep models of Exaone (7.8B and 2.4B) ranked first in all major standards.
• MMLU (understanding a huge multi -task language): The 32B model scored 83.0 on MMLU standards, which LG AI Research is the best performance among local Korean models.

Exaone Deep 32B model has already received international grade.

Shortly after its release, it was included in the list of “artificial intelligence models” by the US -based Research Foundation, Epoch Ai. This list puts the Exaone deep alongside its predecessor, Exaone 3.5, making LG the only Korean entity with the models offered in this prestigious menu in the past two years.

Mathematics ingenuity

Exaone Deep showed exceptional sports thinking skills across their different typical sizes (32B, 7.8B, and 2.4B). In the assessments based on the mathematics curriculum in the academic year 2025, all three models surpassed international thinking models of similar size.

Form of 32B score 94.5 in the general mathematics efficiency test and 90.0 in the American propaganda mathematics exam (AIME) 2024, a qualified test for the American Olympiad.

In AIME 2025, 32B match with Deepseek-R1-Model 671B is much larger. This result shows the effective learning of Exaone Deep and the capabilities of strong logical thinking, especially when dealing with difficult mathematical problems.

Smaller models 7.8B and 2.4B also achieved higher classifications in the main standards of lightweight models and devices, respectively. 7.8B 94.8 recorded the Math-500 and 59.6 standard in AIME 2025, while the 2.4B model achieved 92.3 and 47.9 in the same ratings.

Excellence in science and coding

Exaone Deep also offered great capabilities in professional scientific thinking and programs coding.

Register 32B 66.1 Form in GPQA Diamond test, which holds problem solving skills in physics, chemistry and biology at the PhD level. In the LiveCodebeench evaluation, which measures the efficiency of coding, the model has achieved a degree of 59.5, indicating its capabilities in high -level applications in these areas of experts.

The 7.8B and 2.4B models continued this trend of strong performance, both of which get the first place in GPQA Diamond and LiveCOCOBENCH standards within their size categories. This achievement depends on the success of the EXAONE 3.5 2.4B model, which was previously at the top of the Llm Huging Face in the Edge section.

Rewarded general knowledge

Besides her thinking capabilities, Exaone Deep also showed improved performance in understanding general knowledge.

The 32B model has made a wonderful degree of 83.0 on the MMLU standard, as it placed it as a local model for the best performance in this comprehensive evaluation. This indicates that Exaone Deep thinking exceeds the specific areas and contributes to a broader understanding of various topics.

LG AI Research believes that the developments of thinking about Exaone Deep represent a leap towards a future as Amnesty International can deal with increasingly complex problems and contribute to enriching and simplifying human life through continuous research and innovation.

See also: Baidu undermines artificial intelligence competition models with Ernie 4.5 and Ernie X1

Do you want to learn more about artificial intelligence and large data from industry leaders? Check AI and Big Data Expo, which is held in Amsterdam, California, and London. The comprehensive event was identified with other leading events including the smart automation conference, Blockx, the digital transformation week, and the Cyber ​​Security & Cloud.

Explore the upcoming web events and seminars with which Techforge works here.