K-EXAONE Models by LG Research
1 model2025Up to 256k ctx
Last refreshed 2026-06-29. Next refresh: weekly.
About
LG AI Research's K-EXAONE series of open-source Mixture-of-Experts large language models. K-EXAONE uses a fine-grained MoE architecture and achieved first place in 10 of 13 benchmarks under South Korea's national AI foundation model project. The series supports multilingual use including Korean, English, Spanish, German, Japanese, and Vietnamese.
Decision facts
- Best fit
- General model comparison
- Capability starting point
- K-EXAONE 236B-A23B with 256k context
- Lowest tracked input
- Not tracked
- Closest related family
- Claude 3
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
1 in view
K-EXAONE 236B-A23BCurrent
Use when the workload needs 256k context and 236B parameters.
2025-12256k context236B parameters
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| K-EXAONE 236B-A23B | Use when the workload needs 256k context and 236B parameters. | 2025-12 | 256k context236B parameters | Current |
Release Timeline
1 release group2025-12
1 current
K-EXAONE 236B-A23B
Current256k context236B parameters
Specifications(1 models)
| Model | Released | Context | Parameters |
|---|---|---|---|
| K-EXAONE 236B-A23B | 2025-12 | 256k | 236B |