LG AI Research has open-sourced the lightweight model of its latest AI, EXAONE 3.0, to support the AI research ecosystem. This model, known for its performance and cost-efficiency, is designed for various research applications.
EXAONE 3.0 significantly improves upon its predecessor, EXAONE 2.0, with 56% faster processing, 35% lower memory usage, and 72% reduced operating costs. It has also demonstrated top performance in global benchmarks, excelling in real-world applications, coding, and math, and outperforming similar models like Meta’s Llama 3.1 and Google’s Gemma 2.
Additionally, LG is launching an open beta of ChatEXAONE, an enterprise AI agent based on EXAONE 3.0, to enhance productivity for its employees. This agent offers functions such as real-time web-based Q&A, document analysis, coding assistance, and database management.
LG AI Research has focused on enhancing EXAONE 3.0′s capabilities by using a vast dataset of over 60 million data points from various fields, with plans to expand to over 100 million data points by the end of the year. This comprehensive training is aimed at improving the model’s performance across multiple specialized domains such as law, biology, and education.
LG plans to integrate EXAONE 3.0 into its products and services by the second half of the year, collaborating with its affiliates to fine-tune the model for various business applications, accelerating innovation across the company. The initiative underscores LG’s commitment to advancing AI technology and its practical applications in real-world scenarios.