LG AI Research on Wednesday unveiled the multimodal AI model EXAONE 4.5, which understands and reasons over text and images at the same time.

The model is a preparatory step to expand modalities for K-EXAONE, which is under development in a proprietary AI foundation model project.

LG AI Research plans to move in earnest to expand modalities if its move into the third phase is confirmed after the end of the project’s second phase in August. It ultimately aims to develop EXAONE into physical intelligence that understands and makes judgments about the physical world beyond virtual environments.

According to LG AI Research, EXAONE 4.5 has strengths in accurately reading and reasoning over complex documents used in industrial settings such as contracts, technical drawings, financial statements and scanned documents.

EXAONE 4.5 scored an average of 77.3 points across 5 indicators measuring STEM performance, beating U.S. OpenAI GPT-5 Mini at 73.5 points, Anthropic Claude Sonnet 4.5 at 74.6 points, and China’s Alibaba Qwen3 235B at 77.0 points.

It also outperformed GPT-5 Mini, Claude Sonnet 4.5 and Qwen3-VL on an average score across 13 indicators that include 3 benchmarks measuring general visual understanding and 5 benchmarks evaluating document understanding and reasoning for reading complex information in professional literature, including infographics combining images and text.

In LiveCodeBench v6, a representative coding-performance benchmark, it scored 81.4 points to beat Google’s latest model Gemma 4, which scored 80.0 points. In ChartQA Pro, which evaluates the ability to analyse and reason over complex charts, it scored 62.2 points.

An LG AI Research official explained that a high average score on visual-capability benchmarks means AI has moved beyond simply recognising text or unstructured data in documents to understanding context and answering questions.

LG AI Research released EXAONE 4.5 on the global open-source platform Hugging Face for research, academic and educational use. It also expanded its officially supported languages to Spanish, German, Japanese and Vietnamese, in addition to Korean and English.

Jin-sik Lee (이진식), head of the EXAONE Lab at LG AI Research, said EXAONE 4.5 shows that LG AI has entered a multimodal era in which it understands visual information beyond text. He said the company will start with this model to expand AI’s scope of understanding to speech, video and physical environments, and build AI that makes practical judgments and takes action in industrial settings.