EXAONE 4.5

LG AI Research's first open-weight vision-language model. Integrates a 1.2B vision encoder with the EXAONE 4.0 32B LLM via native multimodal pretraining. Supports 256K token context for long-document and enterprise-scale deployments. Trained with emphasis on document-centric corpora aligned with LG's industrial applications.

Outperforms state-of-the-art models of similar scale on document understanding and Korean contextual reasoning. Achieves 77.3 average across STEM benchmarks, 81.4 on LiveCodeBench v6, 62.2 on ChartQA Pro. Open weights released.