EXAONE Deep: LG AI Research’s Next-Gen Reasoning Model
LG AI Research has unveiled EXAONE Deep, a breakthrough AI model excelling in mathematical reasoning, scientific problem-solving, and coding efficiency.
Introduction
The AI race for advanced reasoning models is heating up, and LG AI Research has made a bold move with the introduction of EXAONE Deep. Competing with leading AI reasoning models, EXAONE Deep demonstrates exceptional performance across mathematics, science, and coding challenges.
Notably, the 32B model has achieved international recognition, making its way into the prestigious ‘Notable AI Models’ list by Epoch AI. Let’s take a closer look at its capabilities.
Mathematical Reasoning Excellence
EXAONE Deep’s reasoning capabilities have set new benchmarks in mathematics. Across different model sizes, it has outperformed global competitors:
- EXAONE Deep 32B: Scored 94.5 in a general mathematics competency test and 90.0 in AIME 2024.
- EXAONE Deep 7.8B: Achieved 94.8 on the MATH-500 benchmark and 59.6 on AIME 2025.
- EXAONE Deep 2.4B: Secured 92.3 on MATH-500 and 47.9 in AIME 2025.
Remarkably, the 32B model matched the performance of DeepSeek-R1 (671B) in the 2025 AIME exam, showcasing its efficiency in logical problem-solving.
Science & Coding Capabilities
EXAONE Deep has also shown excellence in professional science reasoning and coding applications:
- GPQA Diamond Test (doctoral-level physics, chemistry, and biology): Scored 66.1 (32B model).
- LiveCodeBench (software coding proficiency): Achieved a score of 59.5.
The lightweight 7.8B and 2.4B models also ranked first in their categories, continuing the strong performance trend.
Performance Benchmarks
LG AI Research has released the following performance metrics for EXAONE Deep models:
| Benchmark | 32B Model | 7.8B Model | 2.4B Model |
|---|---|---|---|
| MMLU Score | 83.0 | — | — |
| Math General Competency | 94.5 | 94.8 | 92.3 |
| AIME 2025 | 90.0 | 59.6 | 47.9 |
| GPQA Diamond | 66.1 | — | — |
| LiveCodeBench | 59.5 | — | — |
Conclusion
EXAONE Deep has cemented itself as a leading AI reasoning model, excelling in mathematical, scientific, and coding problem-solving. Its performance efficiency, even at smaller sizes, makes it a strong competitor against much larger AI models.
As AI research progresses, LG AI Research’s commitment to improving EXAONE Deep’s capabilities suggests a promising future where AI tackles complex problems across various domains.