whatmodel.sh / EXAONE Deep 32B
EXAONE Deep 32B
↗ share
+ contribute
EXAONE Deep 32B is an open-weights language model from LG AI Research, released March 2025. Its published size is 32B parameters and a 33k-token context window. Licence: exaone.
Catalogue identity notes
LGAI-EXAONE/EXAONE-Deep-32B @4e9a4c2 (created 2025-03-12, EXAONE AI Model License (non-commercial terms)). EXAONE Deep reasoning model (Mar 2025). Lab rows only: rule-(b) rows from the EXAONE Deep card table (all three sizes), pooled with first-party-lgai-scores GPQA Diamond / AIME 2025 runs of the catalogued EXAONE 4.x models.
Model information sources
Release date ↗ · Weights licence ↗
HF safetensors metadata 32,003,200,000 parameters.
$ how it compares
The LLMs ranked either side of EXAONE Deep 32B, out of 828 with a score, then more from LG AI Research.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 294 |  Grok 2 (1212) | 56±1 | comparison only | ||
| 295 |  Qwen3-VL 4B Thinking | 56±1 | local | ||
| 296 | EXAONE Deep 32B | 56±2 | local | ||
| 297 |  DeepSeek R1 Distill Qwen 32B | 56±1 | api $1.59/M +local | ||
| 298 | Qwen2.5-VL 72B Instruct | 56±1 | local | ||
| more from LG AI Research | |||||
| 127 | K-EXAONE 2.0 (750B-A37B) | 67±1 | local | ||
| 162 | K-EXAONE (236B-A23B) | 65±1 | local | ||
| 164 | EXAONE 4.5 33B | 65±2 | local | ||
$ what it's good at
coding██·········█ #1 of 3 * · 11% of its score · 1 test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LiveCodeBench (version unspecified) lab | 59.5% ✎ | ▁···▁▁▃▅▅█▄▅ beats 24/37 | 11% |
math▁▂▃▅▆▅▄█▄▁▁▁ #186 of 398 · 56% of its score · 3 tests
Beats MiniMax-M2.1, DeepSeek V3 0324, Qwen3 235B-A22B (Apr 2025, hybrid) here, though they rank higher overall. Trails Phi-4-mini-reasoning 3.8B, DeepSeek R1 Distill Qwen 7B, OpenThinker3 7B here, though they rank lower overall.
other tests1 not in a task · 32% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| GPQA Diamond lab | 66.1% ✎ | █···▄▄···▄·█ beats 3/6 | 32% |
$ where to run it
llama.cpp · recommended
- 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 | shWindowsirm https://llama.app/install.ps1 | iex - Run EXAONE Deep 32B · needs ~19.34 GB · from huggingface.co/LGAI-EXAONEchat in the terminal
llama cli -hf LGAI-EXAONE/EXAONE-Deep-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-Deep-32B-GGUF:Q4_K_M
LM Studio
- Install LM Studio · Open the app once before using the lms command.
- Run EXAONE Deep 32B · needs ~19.34 GB · from huggingface.co/LGAI-EXAONE
- Or download it in the terminal
lms get https://huggingface.co/LGAI-EXAONE/EXAONE-Deep-32B-GGUF@Q4_K_M -y - Then run lms ls to see its name, and lms chat followed by that name
Ollama
- Install Ollama
download it (macOS / Windows) ↗
Linuxcurl -fsSL https://ollama.com/install.sh | sh - Run EXAONE Deep 32B · needs ~19 GB · from ollama.comchat in the terminal
ollama run exaone-deep:32b
Weights and documentation: huggingface.co/LGAI-EXAONE ↗
$ how the score was worked out · LG AI Research vendor-reporte… 99%
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 |
|---|---|---|
 LG AI Research vendor-reported benchmark scores for its own EXAONE models (Hugging Face model cards) · lab-reported | 5 | 99% |
| prior · pull toward the middle | 1% | |
| weighted average | = 56.1 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json