Artificial intelligence with EXAONE Deep: LG AI Research presents new reasoning AI model – Agentic AI from South Korea – Image: Xpert.Digital
South Korea's AI offensive: EXAONE Deep sets global standards
LG presents EXAONE Deep: Revolutionary open-source agentic AI
LG AI Research has released EXAONE Deep, an advanced reasoning AI model, as open source, bringing South Korea's AI efforts to the global stage. Unveiled in March 2025 at NVIDIA's GTC developer conference, the model is distinguished by its ability to independently formulate and test hypotheses and make autonomous decisions based on them. This innovative AI solution marks the transition to the era of "agentic AI" and positions LG among the few global companies driving this technology forward. With impressive performance in mathematical, scientific, and coding benchmarks, combined with efficient model sizing, EXAONE Deep represents a significant advancement in AI development.
The EXAONE model family and its development
From the beginnings to EXAONE Deep
The foundation for EXAONE Deep was laid in December 2020 with the establishment of LG AI Research. Led by LG Corp Chairman Koo Kwang-mo, the research division was created with the goal of securing LG's long-term future through AI technology. In a leadership meeting, Koo emphasized: “We must develop AI proactively to secure growth drivers for the 2030s.”
The development of the EXAONE model family began with EXAONE 1.0 in December 2021, a “supergiant AI” model with approximately 300 billion parameters. This was followed by EXAONE 2.0 in July 2023 and EXAONE 3.0 in August 2024, the latter representing a significant milestone as South Korea’s first open-source AI model. At the end of 2024, EXAONE 3.5 was released, featuring improved instruction followership and understanding of longer contexts. EXAONE Deep builds upon this development and focuses specifically on reasoning capabilities.
Technical architecture and model variants
EXAONE Deep is based on a decoder-only transformer architecture and is available in three size variants:
- EXAONE Deep-32B: The flagship model with 32 billion parameters and 64 layers, optimized for maximum reasoning performance.
- EXAONE Deep-7.8B: A lightweight version with 7.8 billion parameters and 32 layers, offering 95% of the performance of the 32B model at only 24% of the size.
- EXAONE Deep-2.4B: An on-device model with 2.4 billion parameters and 30 layers that, despite its small size (7.5% of the 32B model), still achieves 86% of the performance.
All models have a maximum context scope of 32,768 tokens, a significant improvement over previous models. The models were primarily trained on reasoning-specific datasets that account for lengthy thought processes, enabling them to understand more complex relationships and draw logical conclusions.
Related to this:
Performance characteristics and benchmark results
Mathematical reasoning and scientific problem solving
EXAONE Deep demonstrates particularly impressive results in mathematical and scientific reasoning tasks. The 32B model achieved 94.5 points in the mathematics section of the South Korean College Entrance Examination (CSAT) and 90.0 points in the 2024 American Invitational Mathematics Examination (AIME), thus outperforming competing models.
In the MATH-500, an index for evaluating mathematical problem-solving skills, it achieved 95.7 points. Particularly noteworthy is that the model achieves this performance with only about 5% of the size of some “giant” models like DeepSeek-R1 (671 billion parameters).
In the area of scientific reasoning, the 32B model scored 66.1 points in the GPQA Diamond test, which assesses problem-solving skills at the doctoral level in physics, chemistry, and biology. These results underscore the model's ability to understand and apply complex scientific concepts.
Coding skills and general language comprehension
EXAONE Deep also demonstrates its strength in coding and problem-solving. In the LiveCodeBench test, which assesses coding capabilities, the 32B model achieved a score of 59.5. This underscores its potential for applications in software development, automation, and other technical fields that require a high degree of computational accuracy.
In general language comprehension, the model achieved the highest MMLU (Massive Multitask Language Understanding) score among Korean models with 83.0 points. This demonstrates that EXAONE Deep performs well not only in specialized reasoning tasks but also in general language comprehension.
Energy efficiency of the smaller models
The performance of the smaller model variants is particularly noteworthy. The 7.8B model achieved 94.8 points in MATH-500 and 59.6 points in AIME 2025, while the 2.4B model scored 92.3 points in MATH-500 and 47.9 points in AIME 2024. These results place the smaller versions of the EXAONE Deep at the top of their respective categories in all major benchmarks.
The community is particularly surprised by the performance of the 2.4B model. A Reddit post notes that this small model even outperforms the significantly larger Gemma3 27B model in certain benchmarks. One user wrote: “I mean – you’re telling me that a 2.4B model (46.6) outperforms the Gemma3 27B (29.7) in the live code benchmark?”
Application potential and significance in the AI market
Areas of application in industry, research and education
LG AI Research expects EXAONE Deep to be used in various fields. The press release states: “EXAONE Deep will not only be widely used in professional fields required by industries of the future, but also in scientific research and educational fields such as physics and chemistry, demonstrating high performance in assessment indicators of specialized fields such as mathematics, science, and coding.”
A particular focus is on the on-device model (2.4B), which, due to its small size, can be used in devices such as smartphones, automobiles, and robotics. Since the data can be processed securely on the device without requiring a connection to external servers, this model offers advantages in terms of data security and the protection of personal data.
Positioning in the global AI competition
With the release of EXAONE Deep, LG is positioning itself in the increasingly competitive global AI market. The South Korean tech company is thus entering into direct competition with major technology companies such as OpenAI, Google DeepMind, and Chinese AI developers like DeepSeek.
An LG AI Research representative stated: “We announced EXAONE Deep approximately one month after participating in the domestic AI industry competitive diagnostics and inspection meeting held at the National Artificial Intelligence Committee in February, where we also indicated our intention to open-source a DeepSeek R1-level model.” The representative added: “The core of LG’s AI technology is maintaining performance while significantly reducing model size.”
At a time when cost-efficient models are receiving significant attention following the rise of China's DeepSeek in reasoning capabilities, LG's approach of developing smaller but powerful models could represent a strategic advantage.
The importance of reasoning AI and “agentic AI”
From knowledge-based AI to reasoning AI
With EXAONE Deep, LG AI Research is making the transition from “knowledge AI” to “reasoning AI”. While traditional AI models are primarily focused on retrieving and providing information, reasoning AIs like EXAONE Deep can independently formulate hypotheses, test them, and make autonomous decisions based on this information.
This capability marks the entry into the era of “Agentic AI”—active AI capable of “thinking” and acting independently. LG AI Research explains: “Agentic AI refers to an active AI that is capable of making autonomous decisions by independently formulating hypotheses and drawing conclusions to verify them.”
The open-source strategy
A key aspect of the EXAONE Deep release is the decision to make the model available as open source. This follows the strategy that began with EXAONE 3.0, South Korea's first open-source AI model.
The open-source strategy allows developers to use and further develop the model without restrictions for research purposes. This could lead to broader application and further development of the technology and strengthen LG's position in the global AI ecosystem.
Kyung-hoon Bae, President of LG AI Research, stated: “We plan to make this highly versatile and lightweight model available as open source so that universities and research institutes can utilize the latest generative AI technology, contributing to the AI research ecosystem and further improving AI competitiveness.”
Related to this:
Future prospects and current developments
ChatEXAONE: The new standard for AI-powered productivity in business
LG plans to collaborate with its subsidiaries in the second half of the year to integrate EXAONE Deep into various products and services. Depending on the application, EXAONE will be available in different model sizes, from an ultra-lightweight model for on-device AI services to a high-performance model for specialized applications.
A concrete example of the practical application of EXAONE technology is ChatEXAONE, an AI agent for businesses based on EXAONE 3.0, which is already available as an open beta version for LG Group employees. ChatEXAONE offers various features to increase work productivity, including real-time web-based question-and-answer systems, document- and image-based question-and-answer systems, coding support, and database management.
Further development of AI expertise within the LG Group
The development of EXAONE Deep is part of a larger AI strategy within the LG Group. LG has already established an in-house AI graduate school to foster customized engineers with a nine-month master's program and an 18-month doctoral program.
Employees taking these courses work on projects that would be difficult for individual subsidiaries to develop. As part of a pilot project, LG Display developed a design technology to fit more pixels onto the same screen using AI, while LG Electronics and LG Innotek explored methods for accurate demand forecasting with AI that will significantly reduce inventory costs.
Why smaller AI models might be the better choice – A look at EXAONE Deep
With the launch of EXAONE Deep, LG AI Research has reached a significant milestone in AI development. As South Korea's first reasoning AI model based on a Foundation Model, it places LG alongside leading global technology companies developing this advanced AI technology. Its impressive performance in mathematical, scientific, and coding benchmarks, combined with efficient model sizing, underscores the potential of this model for various applications.
LG's approach of developing high-performance AI models of relatively small size is particularly noteworthy. While many AI companies are focusing on ever-larger models, EXAONE Deep demonstrates that with intelligent optimization and specialized training, even smaller models can achieve top performance. This could not only offer economic advantages but also enable the deployment of powerful AI models on edge devices.
With the open-source release of EXAONE Deep, LG AI Research is contributing to the global AI research ecosystem and simultaneously strengthening South Korea's position in the international AI competition. It remains to be seen how this technology will be implemented in various LG Group products and services and what innovations it will enable across different industries.
Related to this:
Your global marketing and business development partner
☑️ Our business language is English or German
☑️ NEW: Correspondence in your native language!
Konrad Wolfenstein
I and my team are happy to be available to you as your personal advisor.
You can contact me by filling out the contact form here wolfenstein@xpert.digital:or simply call me at +49 7348 4088 965. My email address is
I'm looking forward to our joint project.
