LG of South Korea has just launched Exaone 4.0, the first hybrid reasoning AI system in the country, capable of generating and evaluating hypotheses.
Exaone 4.0 is an important step in LG's AI research, designed to compete globally with advanced hybrid reasoning technology from countries like the United States, China, and France.
MAIN CONTENT
Exaone 4.0 is the first hybrid reasoning AI system in South Korea, expanding reasoning capabilities and hypothesis evaluation.
Exaone Deep-32B has previously outperformed many international competitors at prestigious math and science competitions.
OpenAI has also launched new AI reasoning models, including a model that integrates image recognition capabilities and solves complex problems.
What is remarkable about LG's Exaone 4.0 in the field of AI?
Exaone 4.0 is developed as South Korea's first hybrid AI model, possessing the ability to both generate and evaluate hypotheses effectively. According to Ms. Lee Jin Sik, Head of EXAONE Lab at LG AI Research, the company will enhance R&D to help Exaone lead the domestic AI market and expand globally.
Exaone must become the pioneering AI model in South Korea, demonstrating its competitiveness in the global market.
Lee Jin Sik, Head of EXAONE Lab, 2025
This model is expected to explore hybrid reasoning potential, combining multiple AI methods to enhance accuracy and the ability to process more complex information.
What AI reasoning models has LG AI Research developed previously?
LG AI Research launched Exaone 1.0 in 2021 and continued to develop Exaone Deep-32B in March 2025 — South Korea's first AI reasoning model with 32 billion parameters, enhancing learning and reasoning capabilities. Experts rated the performance of Exaone Deep equivalent to Deepseek R1 despite having nearly half the parameters.
Exaone Deep-32B scored 90 at the 2024 USA Math Olympiad, surpassing competitors Deepseek and Alibaba.
Performance Report, 2024
In the 2025 CSAT math exam in South Korea, Exaone Deep scored 94.5, demonstrating an advantage in doctoral-level science compared to QwQ-32B (66.1 vs. 63.3). However, weaknesses in language processing and coding still led it to fall short against models from Deepseek and Alibaba.
Why do AI companies emphasize optimization instead of just increasing parameters?
Increasing the number of parameters often helps improve model performance but also significantly increases computational costs and resource usage. Therefore, many AI organizations are seeking ways to optimize accuracy with fewer parameters for greater economic efficiency and convenience.
This trend aligns with modern computational resources and the need to expand AI applications in practice, providing a faster and more stable user experience.
How does Exaone 4.0 compare with international AI reasoning models?
LG stated that Exaone 4.0 has passed numerous performance tests compared to models from the United States, China, and France. Currently, there are several hybrid reasoning systems on the market, such as Claude from Anthropic and Qwen from Alibaba, but very few companies have successfully developed a hybrid model as good as Exaone 4.0.
Exaone 4.0 is expected to be officially released in Seoul on July 22, 2025, creating a breakthrough in the global AI race.
What new AI reasoning models has OpenAI introduced recently?
OpenAI has just announced two new reasoning models called o3 and o4-mini. The o3 model is optimized for solving complex multi-step scientific problems, accompanied by the Codex CLI feature for direct programming via computer terminal.
Meanwhile, while o4-mini is more lightweight, it still maintains performance comparable to o3. Both models have the ability to analyze visual images, processing blurry and detailed images to enhance accuracy in results.
Soon, OpenAI will release an open-source AI reasoning version alongside GPT-5 to expand utility for the development community.
Sam Altman, CEO of OpenAI, 2025
How do new AI reasoning models impact the technology market?
The development of advanced reasoning models enhances natural language processing capabilities, analyzes complex data, and supports humans in fields such as science, engineering, and research. This promotes innovation and expands AI applications across various industries.
Frequently asked questions
What outstanding features does LG's Exaone 4.0 have? This is the first hybrid reasoning AI model in South Korea, capable of generating and evaluating hypotheses, competing with models worldwide. How does Exaone Deep-32B outperform other AI models? Exaone Deep-32B has 32 billion parameters, excelling in math and science exams compared to Deepseek and Alibaba, even with fewer parameters. Why doesn't increasing parameters always yield better results? Increasing parameters makes the model stronger but also increases computational costs; optimizing the model helps save resources and maintain performance. What is special about OpenAI's new AI reasoning models? O3 and o4-mini support solving complex problems, integrate image recognition, and have a highly anticipated open-source version. When and where will Exaone 4.0 be released? This model is announced by LG to launch in Seoul, South Korea on July 22, 2025.
Source: https://tintucbitcoin.com/lg-cong-bo-ai-ly-luan-lai-dau-tien/
Thank you for reading this article!
Please Like, Comment, and Follow TinTucBitcoin to stay updated with the latest news about the cryptocurrency market and not miss any important information!
