LG AI Research unveils its next-generation hybrid AI model EXAONE 4.0, setting a new benchmark for global AI innovation.
EXAONE 4.0 is a hybrid AI that integrates general natural language processing capabilities with the high-level reasoning capabilities proven through EXAONE Deep. It inherits the step-by-step, thought-based problem-solving approach that EXAONE Deep implemented, and advances its reasoning capabilities to excel even in challenging fields such as math, science, and coding.
It also supports Model Context Protocol (MCP) and Function Calling, providing the technical foundation for realizing Agentic AI. Language support has also been expanded to include Spanish in addition to Korean and English, further enhancing global applicability.
Key Takeaways
Integrated generic + inference model, proven world-class performance
Objectivity and fairness are paramount when evaluating the performance of AI models. However, some of the newest models on the market may interpret their results using self-made benchmarks or relatively low-difficulty evaluation sets to highlight their strengths. While this can be effective in highlighting a model's specific capabilities, there is a risk that it may not reflect the level of competitiveness required in the actual global market.
To overcome these limitations, EXAONE conducts evaluations based on the latest and most challenging global benchmarks, and validates its performance through direct comparisons with leading global models, demonstrating its superior performance and competitiveness across complex tasks in multiple languages, not just Korean.
* Key performance indicators :
MMLU-Pro (knowledge and reasoning) : 81.8
LiveCodeBench v6 (coding) : 66.7
GPQA-Diamond(science) : 75.4
AIME 2025 (math) : 85.3
Image 1. EXAONE 4.0 (32B) benchmarked against open-weight models
In an English-based performance assessment aligned to global standards, EXAONE 4.0 validated its performance across a range of challenging benchmarks: World Knowledge, which compares achievements in advanced knowledge across a range of professions and academic disciplines; Math/Coding, the most challenging to assess advanced reasoning performance; Instruction Following, which measures a user's ability to follow instructions; and Tool Use in the era of Agent AI.
It also excelled in the KMMLU-Pro and KMMLU-Redux benchmarks for advanced Korean language proficiency, and demonstrated its multilingual capabilities by expanding its performance to Spanish.
Image 2. EXAONE 4.0 (32B) benchmarked against frontier models
With world-class performance as its foundation, EXAONE 4.0 is designed to be a high-performance solution that can contribute to real work efficiency and productivity improvements even in industries that require complex problem solving. To enable more researchers and developers to experience and utilize this technology, LG AI Research has released EXAONE 4.0 for research and academic purposes through Hugging Face, a global AI open-source platform.
Dual lineup consisting of a professional model (32B) and an on-device model (1.2B)
EXAONE 4.0 supports two models to cater to various uses.
Professional model (32B): Proved expertise by passing 6 nationally recognized certification exams. Suitable for high-difficulty fields such as law, accounting, and medicine.
On-device model (1.2B): Half the size of its predecessor, but with better performance. Can be safely operated without a server connection, making it strong in security and personal information protection.
Image 3. 3. EXAONE 4.0 (1.2B) benchmark results
Expanded usability with commercial Free Licensing for Educational Institutions and Expanded Use via Commercial API
Starting with EXAONE 4.0, the free licensing previously granted for research and academic purposes has been extended to educational institutions. This means that primary and secondary schools, as well as universities, can now freely use EXAONE 4.0 for educational purposes without a separate approval process.
By incorporating EXAONE into curriculum design, hackathons, and hands-on classes, students will have greater opportunities to interact with the latest EXAONE model. This expansion is expected to help students engage more closely with cutting-edge AI technology in a variety of learning environments.
LG AI Research has entered into a strategic partnership with Friendly AI, an official partner of Hugging Face, to launch an EXAONE 4.0 commercial API service that anyone can use without a GPU, paving the way for more companies to easily incorporate EXAONE 4.0 into their work.
Through these strategic collaborations and technology disclosures, LG AI Research is taking the lead in creating a practical AI ecosystem with Korean and international companies.
EXAONE 4.0 is not just a model with improved performance, but a global competitive advantage solution that connects the potential of AI to reality. LG AI Research will continue to collaborate with various partners, challenge new areas, and take the lead in practical and sustainable AI innovations for people and businesses.