[whatmodel.sh](../../README.md) / EXAONE 4.0 32B # EXAONE 4.0 32B LG AI Research · released 2025-07 · 32B params · exaone **score 58.5** ±0.9 · #218 of 732 · from 29 benchmark results across 4 sources Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only. ## What it's good at ● marks EXAONE 4.0 32B among every model on that task or test, worst to best. 
task            vs the others          rank  counts  tests
agent tasks     ▁▁▃▆●▄▅▅▄▄▂▁    #193 of 313     33%      7
coding          ▁▁▁▃▅▇●▇▅▄▃▁   #241 of 469*    6.8%      2
chat & writing  ▁▁▁▃▆▄▆●▆▇▇▃   #200 of 402*    1.7%      1
math            ▁▁▃▅█▆●█▄▂▁▁    #159 of 431     17%      4
agent tasks · #193 of 313 · 33% of its score Beats [o1 (2024-12-17)](o1-2024-12-17.md), [o3-mini](o3-mini.md), [DeepSeek R1 0528](deepseek-r1-0528.md) here, though they rank higher overall. Trails [SmolLM3 3B](smollm3-3b.md), [Jamba 1.5 Mini](jamba-1-5-mini.md), [Mellum2 12B-A2.5B Instruct](mellum2-12b-a2-5b-instruct.md) here, though they rank lower overall. 
test                     result  vs the others in its run    difficulty            counts  why
τ²-bench Telecom          26.1%  ▁·▅●▆▅▂▃▃▂▃▁ beats 16/89    very hard · avg 12%     9.1%  sharp · at its level · noisy · several …
τ²-bench Retail           59.5%  ▁····▂▁▃▆●▃▁ beats 22/44    medium · avg 37%        7.2%  at its level · several runs
τ²-bench Airline          45.6%  ▁▁·▂▁▃▂●█▅▂▃ beats 14/42    hard · avg 29%          5.6%  at its level · several runs
Terminal-Bench Hard        3.6%  ▂▅●▂▂▆▂▂█▅▅▂ beats 6/27     very hard · avg 2.1%    4.8%  sharp · too hard for it · small run
τ-bench Retail (maker)    62.8%  ▁····▃·▂▇●▂▆ beats 13/28    medium · avg 38%        3.3%  at its level · small run
τ-bench Airline (maker)   51.5%  ▁▁▁·▅▄▁█▇●▁▇ beats 22/28    hard · avg 28%          2.4%  at its level · small run
BFCL v3 (maker)           63.9%  ▁▁▁▁▁▃▅█●▅▄▁ beats 117/171  medium · avg 57%        0.9%  blunt · at its level · small run
coding · #241 of 469 · 6.8% of its score Beats [Qwen3.5 4B](qwen3-5-4b.md), [Qwen3-VL 32B Instruct](qwen3-vl-32b-instruct.md), [Solar Open 100B (102B-A12B)](solar-open-100b.md) here, though they rank higher overall. Trails [QwQ-32B-Preview](qwq-32b-preview.md), [InternLM2.5 20B Chat](internlm2-5-20b-chat.md), [Ministral 3 3B Reasoning 2512](ministral-3-3b-reasoning.md) here, though they rank lower overall. 
test                                result  vs the others in its run  difficulty            counts  why
LiveCodeBench v6 (window unspecif…   66.7%  ▂▃▄▄▅█●▆▄▇▄▆ beats 31/67  medium · avg 46%        3.7%  at its level · small run
LiveCodeBench Pro 2025Q2 (Medium)…    4.8%  █··▄▂▂●█▄·▄▆ beats 8/22   very hard · avg 3.3%    3.1%  sharp · small run
chat & writing · #200 of 402 · 1.7% of its score Beats [Gemma 2 9B IT SimPO](gemma-2-9b-simpo.md), [Granite 4.2 8B](granite-4-2-8b.md), [Mistral Small 4 (119B-A6.5B)](mistral-small-4.md) here, though they rank higher overall. Trails [LFM2.5 1.2B Instruct](lfm2-5-1-2b-instruct.md), [LFM2.5 1.2B Thinking](lfm2-5-1-2b-thinking.md), [LFM2.5 8B-A1B](lfm2-5-8b-a1b.md) here, though they rank lower overall. 
test                                result  vs the others in its run  difficulty      counts  why
IFEval (aggregation unspecified) …   83.7%  ▄·▄·▄·▄●···█ beats 4/6    easy · avg 78%    1.7%  typical
math · #159 of 431 · 17% of its score Beats [MiMo-V2.5 (310B-A15B)](mimo-v2-5.md), [MiniMax-M2.7](minimax-m2-7.md), [MiniMax-M2.1](minimax-m2-1.md) here, though they rank higher overall. Trails [Olmo 3 7B Think](olmo-3-7b-think.md), [Qwen3-VL 4B Thinking](qwen3-vl-4b-thinking.md), [Nemotron 3 Nano 30B-A3B](nemotron-3-nano-30b-a3b.md) here, though they rank lower overall. 
test                        result  vs the others in its run  difficulty        counts  why
AIME 2025 (maker)            85.3%  █·▄▄····●··█ beats 4/6    medium · avg 54%    8.3%  sharp
IMO-AnswerBench (maker)      66.1%  ▁▁·▁▁●▄▆█▄▂▃ beats 7/44   medium · avg 41%    4.0%  at its level · small run
HMMT February 2025 (maker)   72.9%  ▁▁▁▃▁▅●▃██▁▄ beats 10/31  medium · avg 51%    2.8%  at its level · small run
HMMT November 2025 (maker)   78.1%  ▁▁····▁●▃█▇▁ beats 5/23   easy · avg 66%      2.3%  small run
other tests (13) 
test                                result  vs the others in its run    difficulty            counts  why
GPQA Diamond (maker)                 75.4%  █···▄▄···●·█ beats 4/6      hard · avg 51%           11%  sharp · at its level
KCL-Essay vanilla                    32.6%  ▁···▁▃▂●▂▃▃▂ beats 11/23    hard · avg 17%          5.5%  at its level
MMLU-Pro (maker)                     81.8%  ▁▁·▁▁▁▃▂▄▅●▁ beats 136/188  medium · avg 62%        4.7%  typical
EuroEval Life in the UK (English …   86.6%  ▂·▂▂·▂·●▃·█▃ beats 5/14     easy · avg 80%          4.0%  typical
EuroEval HellaSwag (English), val…   79.7%  ▂··▂▂·▂▆▆●·▄ beats 11/15    medium · avg 63%        2.9%  at its level
Humanity's Last Exam (maker)         10.6%  ▁▁▁▄●▆▇██▆▁▃ beats 21/111   very hard · avg 4.3%    2.8%  small run
KCL-MCQA vanilla                     36.4%  ▂···█▂●▆██▆▂ beats 8/23     very hard · avg 29%     2.6%  typical
MMLU-Redux (maker)                   92.3%  ▁▁▃▁▂▃▄▅▇▅●▄ beats 46/63    easy · avg 79%          1.8%  small run
MMLU                                 89.9%  ▁▁▁▁▁▂▂▅██▆● beats 208/214  medium · avg 68%        1.8%  noisy · small run
HellaSwag                            65.7%  ▁▁▁▂▂▂▄●█▇▄▂ beats 55/135   medium · avg 63%        1.7%  at its level · small run
IFBench (aggregation unspecified)…   36.0%  ▂▆▂▄●▄▂▄▄▆█▇ beats 23/86    medium · avg 35%        1.5%  at its level · small run
PIQA                                 75.7%  ▁▁▁▂▂▃●█▅▃▁▁ beats 50/123   easy · avg 83%          0.4%  blunt · at its level · small run
Thunder-KoNUBench symbol 0-shot      81.5%  █··········● beats 1/1                                 —  not counted yet: no other model in the …
## How the score was worked out Each result implies a score on its own, after adjusting for how hard that test is and who ran it. The model's 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 is added up by who ran the tests. [Method](../method.md) 
source                                    results  share of the score
LG AI Research vendor-reported benchmar…       18   64% ▇▇▇▇▇▇▇▇
HyperCLOVA X 32B Think                          7   21% ▇▇▇
Korean Canonical Legal Benchmark: Towar…        2    8% ▇
EuroEval leaderboards                           2    7% ▇
prior: pull toward the middle                        0% ▇
weighted average                                   = 58.5
Licences: HyperCLOVA X 32B Think ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) · Korean Canonical Legal Benchmark: Toward Knowledge-Independent Evaluation of LLMs’ Legal Reasoning Capabilities ([CC BY 4.0](../../licenses/cards-final-NOTICE.txt)) · EuroEval leaderboards ([MIT](https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/LICENSE))
Not counted (1) 
source                  test                            score  why
Thunder-KoNUBench: A …  Thunder-KoNUBench symbol 0-sh…  81.5%  too few models have it yet
## How to run it ### llama.cpp (recommended) 1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md) ```sh # macOS / Linux curl -LsSf https://llama.app/install.sh | sh # Windows irm https://llama.app/install.ps1 | iex ``` 2. Run EXAONE 4.0 32B (gguf-q4_k_m, needs ~19.7 GB, from [huggingface.co/LGAI-EXAONE](https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B-GGUF/tree/main)) ```sh llama cli -hf LGAI-EXAONE/EXAONE-4.0-32B-GGUF:Q4_K_M ``` 3. Or start a local server and open the address it prints (http://localhost:9931) ```sh llama serve -hf LGAI-EXAONE/EXAONE-4.0-32B-GGUF:Q4_K_M ``` Other versions: gguf-q8_0 (~38.9 GB) `llama cli -hf LGAI-EXAONE/EXAONE-4.0-32B-GGUF:Q8_0` ### LM Studio 1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download) 2. Run EXAONE 4.0 32B (gguf-q4_k_m, needs ~19.7 GB, from [huggingface.co/LGAI-EXAONE](https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B-GGUF/tree/main)) ```sh lms get https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B-GGUF@Q4_K_M -y ``` 3. Then run lms ls to see its name, and lms chat followed by that name Other versions: gguf-q8_0 (~38.9 GB) `lms get https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B-GGUF@Q8_0 -y` · 4bit (~18 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/EXAONE-4.0-32B-MLX-4bit -y` · 8bit (~34 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/EXAONE-4.0-32B-MLX-8bit -y` ### Ollama 1. Install Ollama. [Download it (macOS / Windows)](https://ollama.com/download) ```sh # Linux curl -fsSL https://ollama.com/install.sh | sh ``` 2. Run EXAONE 4.0 32B (gguf-q4_k_m, needs ~19.7 GB, from [huggingface.co/LGAI-EXAONE](https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B-GGUF/tree/main)) ```sh ollama run hf.co/LGAI-EXAONE/EXAONE-4.0-32B-GGUF:Q4_K_M ``` Other versions: gguf-q8_0 (~38.9 GB) `ollama run hf.co/LGAI-EXAONE/EXAONE-4.0-32B-GGUF:Q8_0` ### MLX (Mac) 1. Install MLX (Mac). Apple silicon Macs only, text only. ```sh # with uv uv tool install mlx-lm # or pip install mlx-lm ``` 2. Run EXAONE 4.0 32B (4bit, needs ~18 GB, Mac only, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/EXAONE-4.0-32B-MLX-4bit/tree/8ecfb4e16c899b9acdf9a20cf4855ae7dc622728)) ```sh mlx_lm.chat --model lmstudio-community/EXAONE-4.0-32B-MLX-4bit ``` 3. Or start an OpenAI-compatible server at http://localhost:8080/v1 ```sh mlx_lm.server --model lmstudio-community/EXAONE-4.0-32B-MLX-4bit ``` Other versions: 8bit (~34 GB, Mac only) `mlx_lm.chat --model lmstudio-community/EXAONE-4.0-32B-MLX-8bit` Model repository and documentation: [huggingface.co/LGAI-EXAONE](https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B) [interactive version](https://whatmodel.sh/model/exaone-4-0-32b) · [JSON](https://whatmodel.sh/api?format=json&all=1)