LG AI Research launches EXAONE 3.5, enhancing GenAI capabilities and productivity while fostering an open AI research ecosystem.
LG AI Research has open-sourced its latest GenAI model, EXAONE 3.5, just four months after EXAONE 3.0. The updated model boasts top performance in global benchmarks, excelling in long text processing, coding, and real-world applications. Unlike its predecessor, LG has released three variations. A lightweight on-device model, a general-purpose version, and a high-performance model for specialized tasks. This step overcomes accessibility challenges and encourages collaboration by allowing researchers worldwide to validate and use the models.
EXAONE 3.5 integrates advanced Retrieval-Augmented Generation (RAG) technology, reducing hallucinations and improving answer accuracy. Its multi-step reasoning capability deconstructs complex questions logically, offering detailed and reliable responses. Capable of processing up to 100 pages of text simultaneously, EXAONE 3.5 addresses the limitations of earlier GenAI tools by supporting long-form content generation and analysis. These improvements position the model as a global leader in usability and reliability.
LG also launched ChatEXAONE, an enterprise AI agent for employees. This tool incorporates features like “Deep,” which provides comprehensive answers by analyzing multi-layered questions, and “Dive,” which customizes search scopes for targeted results. Employees can use ChatEXAONE for document summarization, data analysis, coding, and more, with robust encryption ensuring data security. This solution enhances workplace productivity and efficiency, bridging AI capabilities with everyday tasks.
LG AI Research continues its focus on expanding the AI ecosystem through collaborations with global tech leaders like AWS, NVIDIA, and Google Cloud. By fostering innovation and scalability, LG aims to position EXAONE and its derivatives as flagship models for industrial applications and national AI competitiveness.