K-EXAONE (Reasoning) — benchmark results

Provider: LG AI. Released 2026-01-12. Access: Open.

Unified ELO 1640 ± 32, rank #408 of 1841 rated models, from 55 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA AIME 202590.33Accuracy (%)90.7
AA LiveCodeBench76.83Pass@1 (%)88.3
AA MMLU-Pro83.75Accuracy (%)87.2
AA IFBench64.69Accuracy (%)76.3
AA-LCR55.7Score (self-reported)71.8
AA Humanity's Last Exam13.07Accuracy (%)69.8
AA CritPt1.12Accuracy (%)69.7
AA GPQA Diamond78.28Accuracy (%)69
AA TAU-2 Bench74.27Accuracy (%)66
AA Long Context Reasoning55.67Accuracy (%)65.1
Artificial Analysis Intelligence Index22.1Intelligence Index63.6
Tau3 Banking14.23Success Rate (%)63.5

Interactive version: theaggregate.ai/model?slug=k-exaone-reasoning · How It Works · Data refreshed daily, snapshot 2026-07-25.