Product Launchllmmixture of expertshybrid attentionhuggingface
LG AI Research Releases K-EXAONE Frontier Model
9.1
LG AI Research released K-EXAONE, a 236-billion-parameter foundation model that topped 10 of 13 benchmarks in the first-round evaluation of South Korea’s government-backed national AI foundation model project, earning an average score of 72. Built as a mixture-of-experts with a hybrid-attention design, K-EXAONE reduces memory and compute by about 70% versus Exaone 4.0 and was published as open weights on Hugging Face.
Key Points
- 1Tops benchmarks: K-EXAONE led 10 of 13 tests with a 72 average score
- 2Achieves efficiency: hybrid attention cuts memory and compute by about 70% versus Exaone 4.0
- 3Enables broader access: 236B MoE model activates ~23B params, runs on A100-class GPUs
Scoring Rationale
High practical impact and efficient architecture, with limitation that it remains an incremental frontier model rather than paradigm-shifting.
Sources
Public references used for this report.
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
Used by DS/ML engineers at top companies
High-Value Overnight OrdersEasyDelivered International ShipmentsMediumOn-Time Delivery Rate by CarrierHard
250 free problems · No credit card
See all Logistics & Shipping problems