AI BriefingLG AI Research Unveils EXAONE 3.0 7.8B Instruction Tuned Model
Key point
LG AI Research has open-sourced EXAONE 3.0 7.8B, an expert-grade AI model with outstanding English and Korean performance.
Details
LG AI Research has pre-released EXAONE 3.0 7.8B Instruction Tuned, a model embodying three years of research achievements, as research-purpose open source. This model is based on a Decoder-only Transformer Architecture, featuring 7.8B parameters and 8T tokens of training data.
Key performance metrics are as follows:
- English Performance: Achieved the #1 average score in Real-world Use Cases, and demonstrated superior performance compared to other models in Math and Coding as well.
- Korean Performance: As a Bi-lingual model supporting both Korean and English, it recorded the #1 average score in both Korean Real-world Use Cases and single benchmarks.
- Cost Efficiency: Through model lightweighting technology, inference time was reduced by 56% and cost was cut by 72% compared to EXAONE 2.0. This brings costs down to about 6% of the level compared to EXAONE 1.0.
In addition, for the model's ethical transparency, LG AI Research plans to transparently disclose the results of Red Teaming as well as areas still requiring improvement, and will continue research to advance the AI ecosystem.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Qwen3.5-based Small Decision Model 'Kev' Released with TypeSafe API Compatibility and Performance Metrics