EXAONE 4.5 released
LG's new 33-billion parameter model is now available in three optimized formats for different hardware.
LG AI Research has officially launched EXAONE 4.5, a significant update to its multimodal AI model series. The new model, boasting 33 billion parameters, is now available for download on Hugging Face in three distinct formats: a standard full-precision version, an FP8 quantized version optimized for faster inference with minimal accuracy loss, and a GGUF version designed for efficient local execution on consumer hardware. This multi-format release strategy directly addresses the diverse deployment needs of developers, from cloud servers to edge devices.
This launch positions EXAONE 4.5 as a direct competitor in the crowded mid-tier open-source model market, which includes models like Llama 3 70B and Qwen 2.5 32B. By providing specialized FP8 and GGUF variants, LG is emphasizing practical usability and accessibility, lowering the barrier for researchers and companies to experiment with and integrate a capable Korean-developed multimodal AI. The model is expected to excel in tasks requiring joint understanding of images and text, continuing the strengths of the EXAONE lineage.
- LG AI Research released EXAONE 4.5, a 33-billion parameter multimodal model.
- Available in three formats: standard, FP8 (for efficient inference), and GGUF (for local use).
- Hosted on Hugging Face, providing direct access for developers and researchers.
Why It Matters
Provides a powerful, accessible open-source multimodal AI option, increasing competition and choice for developers building vision-language applications.