Benchmark Almanac

EXAONE 4.0 32B

EXAONE 4.0 32B.

LG AI Research Released July 15th, 2025 Family EXAONE The lab's model card Compare with others
22
Benchmarks with a score
0
Ranked first, out of five or more models
0
In the top three
LG AI Research Jul 2025

EXAONE 4.0 32B

Open weights Flagship

137.1 #207 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
  • KMMLU-Pro 67.7% #4 /7
  • LiveCodeBench v5 72.6% #6 /47
  • tau-bench 57.1% #17 /23

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
MMLU-Pro 81.8% #62 of 167 Lab model card
MMMLU 85.6% #30 of 71 Lab model card
Multi-IF 73.5% #12 of 28 Lab model card
MMLU-Redux 92.3% #21 of 45 Lab model card
RULER 88.2% #23 of 42 Lab model card
KMMLU-Pro 67.7% #4 of 7 Lab model card
IFEval 84.8% #83 of 135 Lab model card
Global PIQA 51.1% #42 of 44 Official leaderboard
Ko-LongBench (LG) 76.9% #3 of 3 Lab model card
LongBench 48.1% #2 of 2 Lab model card
HELMET 58.3% #1 of 1 Lab model card
KMMLU-Redux 72.7% #1 of 1 Lab model card
LongBench English 48.1% #1 of 1 Lab model card

Math and science

BenchmarkScoreRankSource
GPQA Diamond 75.4% #147 of 302 Lab model card
AIME 2025 85.3% #71 of 143 Lab model card
HMMT 2025 72.9% #46 of 54 Lab model card
HRM8K-KSM 87.6% #2 of 2 Lab model card

Coding

BenchmarkScoreRankSource
LiveCodeBench v5 72.6% #6 of 47 Lab model card
LiveCodeBench 85.9% #8 of 21 Official leaderboard
LiveCodeBench v6 66.7% #65 of 90 Lab model card

Agents and tools

BenchmarkScoreRankSource
Berkeley Function Calling Leaderboard (BFCL) 65.2% #27 of 41 Lab model card
tau-bench 57.1% #17 of 23 Lab model card

Its lab's numbers put it at 137.1, number 207 of 404

The Self-Reported Index fits one score to the 13 capability results LG AI Research reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 135.1 to 139.2. How we compute it.

Its lab's numbers next to 3 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark Lab says Board says Gap
RULER 88.2% RULER (non-reasoning) 93.1% Leaderboard −4.9 pts
LiveCodeBench v6 66.7% LiveCodeBench v6 (reasoning) 70% Leaderboard −3.3 pts
LiveCodeBench v5 72.6% LiveCodeBench v5 (reasoning) 71.7% Leaderboard +0.9 pts

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

EXAONE-4.0-32B model card (Hugging Face) Jul 15, 2025

Its system card

Every result in the longer documents the lab published with it. 2 of them aren't in the launch post.