ModelBeat
Models/K-EXAONE
All models
L

LG AI Research: K-EXAONE

k-exaone

K-EXAONE is LG AI Research’s open-weights AI model, released Dec 31, 2025.

Context
Input
/ 1M
Output
/ 1M
Released
Dec 31, 2025
Modalities

Benchmarks

Scores on standardized evaluations. Higher is better — and rank shows where K-EXAONE lands among all models tracked on Model Beat.

34.0
Intelligence Index
Artificial Analysis
34th percentile of tracked models
28.0
Coding Index
Artificial Analysis
28th percentile of tracked models
24.0
Agentic Index
Artificial Analysis
24th percentile of tracked models

Reasoning

3 evals
GPQA Diamond78.3%

Graduate-level scientific reasoning

Humanity's Last Exam13.9%

Frontier of human expert knowledge

MMLU-Pro83.8%

Broad expert knowledge

Coding

2 evals
LiveCodeBench76.8%

Contamination-free coding

SciCode35.6%

Scientific research coding

Math

1 evals
AIME 2024/202590.3%

Olympiad-qualifier math

Agentic & Tools

1 evals
τ²-bench74.3%

Tool-agent-user reliability

Changelog

1
  • K-EXAONE: Humanity's Last Exam improved from 13.1% to 13.9% (+6%)benchmark · Aug 6, 2026

Compare K-EXAONE with

Frequently asked questions

What is K-EXAONE?

K-EXAONE is an AI model, from LG AI Research, released Dec 31, 2025.

Who created K-EXAONE?

K-EXAONE is an AI model developed by LG AI Research, released Dec 31, 2025.

How does K-EXAONE perform on benchmarks?

On standardized evaluations tracked by Epoch AI, K-EXAONE scores 90.3% on AIME 2024/2025, 83.8% on MMLU-Pro, 76.8% on LiveCodeBench.

Is K-EXAONE open source?

K-EXAONE is released with open weights (Open weights (restricted use)), so the model can be downloaded and self-hosted.

Benchmark data from Epoch AI (CC BY) and Artificial Analysis; pricing, specs & descriptions from OpenRouter. Data updated Sep 17, 2026.