EXAONE 3.5, announced today by LG AI Research is a significant advancement in artificial intelligence technology and a new chapter in the democratization of AI technology. But having just rolled out its predecessor back in April, this release from LG is sure to be their attempt at innovation in the AI landscape within the shortest time possible.
The new EXAONE 3.5 stands out by offering three distinct models tailored for different use cases: We present an ultra light-weight model with 2.4 billion for on device applications, a general purpose model of 7.8 billion parameters, and a particularly powerful model of 32 billion parameters to specialized tasks.
Breaking New Ground in AI Capabilities
The distinguishing feature of EXAONE 3.5 is its usage of advanced Retrieval-Augmented Generation (RAG) that helps the system create more accurate and robust responses by feeding the web search results and document analysis in real time. With this mode of Multi step Reasoning the model transforms into a sophisticated problem solver, capable of carrying out step by step logical analysis.
The most impressive feature of EXAONE 3.5 is that it can handle large text input extensive documents like about 100 pages in a single session. This capacity makes the system a powerful tool for enterprise level document analysis and content generation.
Setting New Industry Standards
Benchmark evaluations show EXAONE 3.5 performs world class for 20 different benchmarks, with best performance on real world text processing and math, as well as coding. LG AI Research publishes detailed performance metrics for all benchmarks, and operates with an open process.
Fostering innovation in the AI community here at LG, we decided to make EXAONE 3.5 open source. All three models are available for research purpose to the developers and researchers to validate and build the technology.
Enterprise Innovation Through ChatEXAONE
Handing along the open source release, LG is also releasing ChatEXAONE, an enterprise AI agent built upon EXAONE 3.5. But first, it comes with robust security features to keep company data safe as employees use AI capabilities for their day to day tasks.
This is where ChatEXAONE introduces new features like ‘Deep’ and ‘Dive’, which improve the way users work with AI. ‘Deep’ feature allows multi stage analysis of complex queries and ‘Dive’ allows users define their scope of search for academic content and YouTube.
Future-Focused Development
With EXAONE 3.5, LG AI Research isn’t resting on its laurel. They’ve already started working to take the technology further by developing Large Action Models (LAM) that can plan and act on their own. By even 2025, the goal of this initiative is to create more sophisticated AI agents.
EXAONE 3.5’s robust capabilities have won partnerships from industry leaders like AWS, Dell Technologies, Google Cloud, and many academic institutions. The idea here is to use these collaborations to actually expand the practical application of the technology across different industries.
Final Thought
EXAONE 3.5’s advanced AI technology being released by LG is a major breakthrough for the research and enterprise use of this technology. The security, reliability, and high standards for which the platform operates offer users a powerful feature set in an open source form factor that forms the basis for innovation in the future of the AI field.
Our agility to leapfrog competition in the AI world is illustrated by LG’s move from EXAONE 3.0 to EXAONE 3.5 in just four months. EXAONE’s evolution will be a key part of what defines the future of AI technology, as the company itself continues to build new capabilities and build strategic partnerships.
