LG AI releases EXAONE 4.5 33B

LG is quietly becoming one of the more important open-weight labs outside the usual US-China axis. The real test is whether EXAONE-4.5-33B becomes a genuinely useful multimodal model or just another respectable 33B that gets benchmarked and forgotten.

• –33B is a practical size for self-hosted inference, so quantizations, serving support, and memory footprint will matter immediately.
• –EXAONE’s move toward vision-language capability broadens the model beyond plain chat and code into more enterprise-friendly workflows.
• –The release adds more Korean and broader Asian-language coverage to the open model landscape, which is still unevenly served.
• –Adoption will hinge on ecosystem support from Transformers, vLLM, TensorRT-LLM, and community GGUF/AWQ ports.