LG AI Research/Mixture of Experts

LK-EXAONE 2.0

Name: K-EXAONE 2.0
Author: LG AI Research

K-EXAONE 2.0 — LG AI Research's 750B MoE with 37B active, Korea's largest open model, under Apache 2.0. Three layers in four use a 128-token sliding window, so only 20 of its 78 layers keep a full KV cache.

chatcodingreasoningmultilingualmathagentictool_use

749.36B

Parameters (37B active)

256K

Context length

Benchmarks

Quantizations

HF downloads

Architecture

MoE

Released

2026-07-31

Layers

KV Heads

Head Dim

128

Family

exaone

Quantization Options

Context length:

Quant	Bits	VRAM @ 16K	Quality
IQ2_XXS	2.38	224.4 GB 223.4 + 0.9 KV	low
IQ2_M	2.93	275.9 GB 274.9 + 0.9 KV	low
Q2_K	3.16	297.4 GB 296.5 + 0.9 KV	low
IQ3_XXS	3.25	305.9 GB 304.9 + 0.9 KV	low
IQ3_XS	3.5	329.3 GB 328.3 + 0.9 KV	low
Q3_K_S	3.64	342.4 GB 341.4 + 0.9 KV	low
IQ3_M	3.76	353.6 GB 352.7 + 0.9 KV	low
Q3_K_M	4	376.1 GB 375.2 + 0.9 KV	low
Q3_K_L	4.3	404.2 GB 403.3 + 0.9 KV	moderate
IQ4_XS	4.46	419.2 GB 418.3 + 0.9 KV	moderate
Q4_K_S	4.67	438.9 GB 437.9 + 0.9 KV	moderate
Q4_K_M	4.89	459.5 GB 458.5 + 0.9 KV	good
Q5_K_S	5.57	523.2 GB 522.2 + 0.9 KV	good
Q5_K_M	5.7	535.3 GB 534.4 + 0.9 KV	good
Q6_K	6.56	615.9 GB 615.0 + 0.9 KV	excellent
Q8_0	8.5	797.6 GB 796.7 + 0.9 KV	lossless
FP16	16	1500.1 GB 1499.2 + 0.9 KV	lossless

Select your GPU above to see speed estimates and compatibility for each quantization.

Too big for a single GPU — plan a multi-GPU deployment

Even the lightest quant needs ~224 GB. Size GPUs, replicas, TCO and scaling for a production setup. Open in Enterprise →

▸ READY TO RUN THIS?RENT BY THE HOUR

RENT A GPU AND RUN K-EXAONE 2.0 NOW

Rent on Vast.ai →Or RunPod →

Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.

Community Ratings

Loading ratings...

Benchmarks (8)

AIME92.3

MMLU-PRO83.5

GPQA Diamond82.9

SWE-bench68.2

AA Long Context60.0

SciCode42.0

Terminal-Bench 2.140.4

HLE18.6

Run this model

▸Easiest way to get started·Beginners

DOCS ↗

curl -fsSL https://ollama.com/install.sh | sh

$ollama run exaone:749b-q4_K_M

Tag may need adjustment — check ollama.com/library/exaone for available tags.

▸ SETUP GUIDE

Auto-setup with fitmyllm CLI

Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.

pip install fitmyllmthen run fitmyllmLearn more

Auto-detect GPULive tok/s in chatSpeed benchmarks9 inference engines

HuggingFace GGUF Downloads Build Hardware

Find the best GPU for K-EXAONE 2.0

Build Hardware for K-EXAONE 2.0

QUANT	BPW	VRAM	QUALITY
IQ2_XXS	2.38	223.4 GB	65%
IQ2_M	2.93	274.9 GB	75%
Q2_K	3.16	296.5 GB	78%
IQ3_XXS	3.25	304.9 GB	82%
IQ3_XS	3.5	328.3 GB	84%
Q3_K_S	3.64	341.4 GB	85%
IQ3_M	3.76	352.7 GB	86%
Q3_K_M	4	375.2 GB	88%
Q3_K_L	4.3	403.3 GB	90%
IQ4_XS	4.46	418.3 GB	92%
Q4_K_S	4.67	437.9 GB	93%
Q4_K_M	4.89	458.5 GB	94%
Q5_K_S	5.57	522.2 GB	96%
Q5_K_M	5.7	534.4 GB	96%
Q6_K	6.56	615.0 GB	97%
Q8_0	8.5	796.7 GB	100%
FP16	16	1499.2 GB	100%