SEOUL (ANN/KOREA HERALD) – LG Group’s artificial intelligence think tank, LG AI Research, has introduced the third generation of its hyperscale AI multimodal model, Exaone 3.0, which offers significant improvements in performance and cost-efficiency.
To boost the AI ecosystem, LG AI Research announced the release of a lightweight model of Exaone 3.0 as Korea’s first open-source AI model for research purposes.
Since its initial launch in December 2021, Exaone has undergone substantial upgrades. The latest version, Exaone 3.0, unveiled on Wednesday, requires 56 per cent less processing time, 35 per cent less memory, and 72 per cent lower operating costs compared to its predecessor.
Addressing the issue of AI power consumption, LG AI Research has managed to reduce the model size by 97 per cent while enhancing performance over Exaone 1.0. Exaone 3.0 has achieved top global rankings in 13 benchmark scores, surpassing competitors like Meta’s Llama 3.1 and Google’s Gemma 2, especially excelling in real-world applications, coding, and math.
The bilingual capabilities of Exaone 3.0, proficient in both Korean and English, have set new standards in Korean language performance. Trained on over 60 million data points from various specialised fields, the model aims to incorporate additional data from law, biology, medicine, education, and other languages, exceeding 100 million data points by year-end.
In collaboration with LG affiliates, LG AI Research plans to integrate Exaone 3.0 into products and services in the latter half of the year and is actively seeking global partnerships. The model’s versatility is highlighted by its diversified sizes, catering to various applications from ultra-lightweight on-device AI to high-performance specialised models.
“As it is important to create AI that can be used in real-world industries, we plan to strengthen partnerships between LG affiliates, other companies, and institutions with Exaone, which has specialised performance and cost efficiency,” said LG AI Research President Bae Kyung-hoon. “We plan to open-source this highly versatile and lightweight model to contribute to the AI research ecosystem and enhance AI competitiveness.”
LG Group has initiated an open beta service of ChatExaone, an enterprise AI agent based on Exaone 3.0, for its affiliates’ employees. This service aims to enhance work productivity through real-time web-based Q&A, document and image-based Q&A, coding, and database management, among other tasks.
LG AI Research plans to release a public beta service by year-end, incorporating user feedback, and will roll out official services and mobile apps as readiness permits.
Since its establishment in December 2020, LG AI Research has been accelerating its AI transformation, planning to invest KRW3.6 trillion (USD2.84 billion) in AI and data R&D by 2026. LG Group Chairman Koo Kwang-mo has identified AI as a key growth engine, emphasising bold investment and innovation, leading to significant advancements in AI technology applications and external partnerships.
