K-EXAONE 236B-A23B
Released 2026-01 · hybrid · 260K tokens · 6 benchmarks · Open weight
Editorial notes
Lanzado 12 enero 2026. MoE 236B totales / 23B activos. Hybrid Attention (sliding window + global), -70% memoria vs EXAONE 4.0. Primer modelo no-US/no-China en Top 10 global open-weight. Lidera 10/13 benchmarks del proyecto nacional Korea (promedio 72).
Spec sheet
- Empresa
- LG AI Research
- Pais
- KR
- Tipo
- hybrid
- Release
- 2026-01
- Context
- 260K tokens
- Params total
- 236B
- Params active (MoE)
- 23B
- Licencia
- k-exaone-license
- Quants
- BF16, Q8_0, Q5_K_M, Q4_K_M
- Slug
- k-exaone-236b
Benchmarks (6)
Reasoning 3
Cite this model
BibTeX · APA
BibTeX
@misc{frontier-k-exaone-236b,
title = {K-EXAONE 236B-A23B},
author = {{LG AI Research}},
year = {2026},
note = {Frontier Benchmarks AI atlas. Accessed 2026-05-08},
url = {https://frontierbenchmarks.com/models/k-exaone-236b}
} APA
LG AI Research (2026). K-EXAONE 236B-A23B [Large language model]. Frontier Benchmarks AI. Retrieved 2026-05-08, from https://frontierbenchmarks.com/models/k-exaone-236b
Citation refleja la pagina del atlas, no el paper original del modelo. Para el paper, ve a la seccion "Recursos" arriba.