K-EXAONE 2.0: benchmark results
LG AI Research's 750B mixture-of-experts model (37B active), released 31 July 2026 under Apache 2.0 as Korea's largest open foundation model, built under the Ministry of Science and ICT's Sovereign AI Foundation Model Project. Provider: LG AI. Released 2026-07-31. Access: Open.
Aggregate Rating 1660 ± 35, rank #298 of 1635 rated models, from 18 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Korean CSAT 2026 (Easy Mode) - Mathematics | 100 | Points (out of 100) | 88.7 |
| AA GPQA Diamond | 82.93 | Accuracy (%) | 72.5 |
| AA Humanity's Last Exam | 18.58 | Accuracy (%) | 63.7 |
| Arcee MMLU-Pro (BenchLM) | 83.5 | Accuracy (%) | 61.6 |
| AA CritPt | 0.86 | Accuracy (%) | 54.3 |
| AA Long Context Reasoning | 60 | Accuracy (%) | 49 |
| Korean CSAT 2026 (Easy Mode) - Total | 392 | Points (out of 450) | 46 |
| Korean CSAT 2026 (Easy Mode) - Chemistry I | 41 | Points (out of 50) | 45 |
| Korean CSAT 2026 (Easy Mode) - Society and Culture | 34 | Points (out of 50) | 42.7 |
| AA Terminal-Bench 2.1 | 40.45 | Pass@1 (%) | 41.5 |
| Korean CSAT 2026 (Easy Mode) - Korean | 88 | Points (out of 100) | 41.3 |
| Korean CSAT 2026 (Easy Mode) - English | 94 | Points (out of 100) | 38.7 |
Interactive version: theaggregate.ai/model?slug=k-exaone-2-0 · How It Works · Data refreshed daily, snapshot 2026-11-05.