K-EXAONE — local AI model by LG AI Research
Written by Jakub Rusinowski · Last updated
LG AI Research's frontier-scale model, developed for Korea's sovereign-AI programme: K-EXAONE 2.0 is a 750B mixture-of-experts with 37B active, released under Apache-2.0, unlike the non-commercial EXAONE 4 line. A multi-GPU server model.
Variants
The smallest K-EXAONE variant needs about 454 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache.
| Model | VRAM at Q4 | VRAM | Context | Run it |
|---|---|---|---|---|
| K-EXAONE 2.0 750B-A37B → 750B (37B active) | ~453.6 GB | 262,144 | ollama run k-exaone |
Memory is quantized weights plus overhead at Q4_K_M, from the same engine as the GPU & VRAM checker.
How to run K-EXAONE locally
Install Ollama, then pull the tag.
ollama run k-exaonePick a size above for its own VRAM figure, speed estimate and install command.
Licence
Commercial use permitted. No usage restrictions beyond attribution.
Applies to: K-EXAONE 2.0 750B-A37BK-EXAONE — frequently asked questions
How much VRAM does K-EXAONE need?
K-EXAONE needs about 454 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: K-EXAONE 2.0 750B-A37B (454 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run K-EXAONE on an RTX 4090 (24 GB)?
K-EXAONE's smallest variant needs about 454 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
What quantization should I use for K-EXAONE?
Q4_K_M is the best balance of quality and VRAM for K-EXAONE in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run K-EXAONE with Ollama?
K-EXAONE has no local Ollama tag — the published tag is cloud-hosted, so running it sends your prompts to a hosted GPU rather than your own machine.