EXAONE 4.0 32B

+ contribute

EXAONE 4.0 32B is an open-weights language model from LG AI Research, released July 2025. Its published size is 32B parameters and a 131k-token context window. Licence: exaone.

58.7±0.9#209 of 719 LLMsfrom 29 benchmark results across 4 sources

$ how it compares

The LLMs ranked either side of EXAONE 4.0 32B, out of 828 with a score, then more from LG AI Research.

#	model	score	where	RAM	released
249	Trinity-Large-Preview Arcee AI	59±2	local	~155GB	2026-01
250	GLM-4.7-Flash Z.ai	59±1	api $0.0000/M +local	~19GB	2026-01
251	EXAONE 4.0 32B LG AI Research	59±1	local	~20GB	2025-07
252	Phi-4-reasoning-plus 14B Microsoft	59±1	local	~10GB	2025-04
253	Nemotron-Cascade-8B NVIDIA	59±1	local	~6.3GB	2025-12
	more from LG AI Research
127	K-EXAONE 2.0 (750B-A37B) LG AI Research	67±1	local	~791GB	2026-07
162	K-EXAONE (236B-A23B) LG AI Research	65±1	local	~152GB	2025-12
164	EXAONE 4.5 33B LG AI Research	65±2	local	~24GB	2026-04

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #181 of 300 · 32% of its score · 7 tests

Beats o1 (2024-12-17), o3-mini, DeepSeek R1 0528 here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Mini, Mellum2 12B-A2.5B Instruct here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
τ²-bench Telecom +1	26.1% ✎	▁·▅█▆▅▂▃▃▂▃▁ beats 16/89	hard · avg 17%	▇▇▇▇ 8.9%	sharp · at its level · small run · several runs
τ²-bench Retail +1	59.5% ✎	▁····▂▁▃▆█▃▁ beats 22/44	medium · avg 38%	▇▇▇ 6.8%	at its level · several runs
τ²-bench Airline +1	45.6% ✎	▁▁·▂▁▃▂▃█▅▂▃ beats 14/42	hard · avg 29%	▇▇ 5.2%	at its level · several runs
Terminal-Bench Hard	3.6% ✎	▂▅▆▂▂▆▂▂█▅▅▂ beats 6/27	very hard · avg 2.3%	▇▇ 4.6%	sharp · too hard for it · small run
τ-bench Retail lab	62.8% ✎	▁····▃·▂▇█▂▆ beats 13/28	medium · avg 38%	▇ 3.6%	at its level · small run
τ-bench Airline lab	51.5% ✎	▁▁▁·▅▄▁█▇▃▁▇ beats 22/28	hard · avg 27%	▇ 2.6%	at its level · small run
BFCL v3 lab	63.9% ✎	▁▁▁▁▁▃▅█▆▅▄▁ beats 117/170	medium · avg 57%	▇ 0.9%	blunt · at its level · small run

coding▁▁▁▃▆██▇▆▄▃▁ #229 of 458 * · 7.4% of its score · 2 tests

Beats Qwen3.5 4B, Qwen3-VL 32B Instruct, Solar Open 100B (102B-A12B) here, though they rank higher overall. Trails QwQ-32B-Preview, Ministral 3 3B Reasoning 2512, GPT-4 (0613) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LiveCodeBench v6 (window unspecified) lab	66.7% ✎	▂▃▄▄▅██▅▄▇▃▆ beats 31/65	medium · avg 46%	▇▇ 3.8%	at its level · small run
LiveCodeBench Pro 2025Q2 (Medium) lab	4.8% ✎	█··▄▂▂██▄·▄▆ beats 8/22	very hard · avg 3.0%	▇ 3.5%	sharp · small run

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #194 of 393 * · 1.6% of its score · 1 test

Beats Granite 4.2 8B, Qwen3-VL 30B-A3B Thinking, Mistral Small 4 (119B-A6.5B) here, though they rank higher overall. Trails LFM2.5 1.2B Instruct, LFM2.5 1.2B Thinking, Gemma 3 12B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
IFEval (aggregation unspecified) lab	83.7% ✎	▄·▄·▄·▄▄···█ beats 4/6	easy · avg 78%	▇ 1.6%	typical

math▁▂▃▅▆▅▄█▄▁▁▁ #151 of 398 · 18% of its score · 4 tests

Beats MiMo-V2.5 (310B-A15B), MiniMax-M2.7, MiniMax-M2.1 here, though they rank higher overall. Trails Nemotron 3 Nano 30B-A3B, Qwen3-VL 8B Thinking, Olmo 3 32B Think here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
AIME 2025 lab	85.3% ✎	█·▄▄····▄··█ beats 4/6	medium · avg 53%	▇▇▇ 7.8%	sharp
IMO-AnswerBench lab	66.1% ✎	▁▁·▁▁▃▄▅█▄▂▃ beats 6/42	medium · avg 40%	▇▇ 4.2%	at its level · small run
HMMT February 2025 lab	72.9% ✎	▁▁▁▃▁▅▅▃██▁▄ beats 10/31	medium · avg 51%	▇ 3.3%	at its level · small run
HMMT November 2025 lab	78.1% ✎	▁▁····▁▅▃█▇▁ beats 5/23	easy · avg 65%	▇ 2.7%	small run

other tests13 not in a task · 40% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond lab	75.4% ✎	█···▄▄···▄·█ beats 4/6	hard · avg 51%	▇▇▇▇ 11%	sharp · at its level
KCL-Essay vanilla	32.6% ✎	▁···▁▃▂█▂▃▃▂ beats 11/23	hard · avg 18%	▇▇ 5.0%	at its level
MMLU-Pro lab	81.8% ✎	▁▁·▁▁▁▃▂▄▅█▁ beats 136/188	medium · avg 62%	▇▇ 4.5%	at its level
EuroEval Life in the UK (English knowledge MC), validation split, 5-shot, accuracy	86.6% ✎	▂·▂▂·▂·▃▃·█▃ beats 5/14	easy · avg 80%	▇▇ 4.3%	typical
EuroEval HellaSwag (English), validation split, 5-shot generative MC, accuracy	79.7% ✎	▂··▂▂·▂▆▆█·▄ beats 11/15	medium · avg 63%	▇ 3.0%	at its level
Humanity's Last Exam lab	10.6% ✎	▁▁▁▃▆▆▇██▆▁▃ beats 18/104	very hard · avg 4.4%	▇ 2.7%	small run
KCL-MCQA vanilla	36.4% ✎	▂···█▂▆▆██▆▂ beats 8/23	very hard · avg 30%	▇ 2.4%	typical
HellaSwag	65.7% ✎	▁▁▁▁▂▂▃▇█▆▅▃ beats 31/85	medium · avg 64%	▇ 2.3%	at its level · small run

5 more tests

test	result	vs the others in its run	difficulty	counts	why
MMLU-Redux lab	92.3% ✎	▁▁▃▁▂▃▄▅▇▅█▄ beats 46/63	easy · avg 79%	▇ 1.9%	small run
IFBench (aggregation unspecified) lab	36.0% ✎	▃▇▂▄▅▅▂▄▄▆█▇ beats 23/80	medium · avg 36%	▇ 1.6%	at its level · small run
MMLU	89.9% ✎	▁▁▁▁▁▂▃▅█▃▆▃ beats 177/183	medium · avg 71%	▇ 1.5%	noisy · small run
PIQA	75.7% ✎	▁▁▁▃▃▄▇█▄▃▁▁ beats 38/83	easy · avg 84%	▇ 0.6%	blunt · small run
Thunder-KoNUBench symbol 0-shot	81.5% ✎	█··········█ beats 1/1		—	not counted yet: no other model in the same run to compare it with

$ where to run it

llama.cpp · recommended

version

Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run EXAONE 4.0 32B · needs ~19.7 GB · from huggingface.co/LGAI-EXAONE
chat in the terminal
llama cli -hf LGAI-EXAONE/EXAONE-4.0-32B-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf LGAI-EXAONE/EXAONE-4.0-32B-GGUF:Q4_K_M

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run EXAONE 4.0 32B · needs ~19.7 GB · from huggingface.co/LGAI-EXAONE
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B-GGUF@Q4_K_M -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run EXAONE 4.0 32B · needs ~19.7 GB · from huggingface.co/LGAI-EXAONE
chat in the terminal
ollama run hf.co/LGAI-EXAONE/EXAONE-4.0-32B-GGUF:Q4_K_M

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run EXAONE 4.0 32B · needs ~18 GB, Mac only · from huggingface.co/lmstudio-community
chat in the terminal
mlx_lm.chat --model lmstudio-community/EXAONE-4.0-32B-MLX-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model lmstudio-community/EXAONE-4.0-32B-MLX-4bit

Weights and documentation: huggingface.co/LGAI-EXAONE ↗

$ how the score was worked out · LG AI Research vendor-reporte… 65%, HyperCLOVA X 32B Think (NAVER… 20%, Korean Canonical Legal Benchm… 7% and 1 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
LG AI Research vendor-reported benchmark scores for its own EXAONE models (Hugging Face model cards) · lab-reported	18	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 65%	Factual vendor-reported score (not licensed content)
HyperCLOVA X 32B Think (NAVER Cloud, arXiv:2601.03286v1)	7	▇▇▇▇▇▇ 20%	CC-BY-4.0
Korean Canonical Legal Benchmark: Toward Knowledge-Independent Evaluation of LLMs’ Legal Reasoning Capabilities	2	▇▇ 7%	CC BY 4.0
EuroEval leaderboards (results collection), English generative leaderboard	2	▇▇ 7%	MIT
prior · pull toward the middle		▇ 0%
weighted average		= 58.7