SK Telecom/Mixture of Experts

SA.X K2

Name: A.X K2
Author: SK Telecom

A.X K2 — SK Telecom's 688B MoE with 33B active, trained from scratch natively in FP8 for Korea's sovereign-AI programme. MLA attention with a sparse top-k indexer, and a single model that switches between thinking and non-thinking modes.

chatcodingreasoningmultilingualmathagentictool_use

691.69B

Parameters (33B active)

256K

Context length

Benchmarks

Quantizations

HF downloads

Architecture

MoE

Released

2026-07-30

Layers

KV Heads

Head Dim

112

Family

other

Quantization Options

Context length:

Quant	Bits	VRAM @ 16K	Quality
IQ2_XXS	2.38	207.1 GB 206.3 + 0.8 KV	low
IQ2_M	2.93	254.6 GB 253.8 + 0.8 KV	low
Q2_K	3.16	274.5 GB 273.7 + 0.8 KV	low
IQ3_XXS	3.25	282.3 GB 281.5 + 0.8 KV	low
IQ3_XS	3.5	303.9 GB 303.1 + 0.8 KV	low
Q3_K_S	3.64	316.0 GB 315.2 + 0.8 KV	low
IQ3_M	3.76	326.4 GB 325.6 + 0.8 KV	low
Q3_K_M	4	347.1 GB 346.3 + 0.8 KV	low
Q3_K_L	4.3	373.1 GB 372.3 + 0.8 KV	moderate
IQ4_XS	4.46	386.9 GB 386.1 + 0.8 KV	moderate
Q4_K_S	4.67	405.1 GB 404.3 + 0.8 KV	moderate
Q4_K_M	4.89	424.1 GB 423.3 + 0.8 KV	good
Q5_K_S	5.57	482.9 GB 482.1 + 0.8 KV	good
Q5_K_M	5.7	494.1 GB 493.3 + 0.8 KV	good
Q6_K	6.56	568.5 GB 567.7 + 0.8 KV	excellent
Q8_0	8.5	736.2 GB 735.4 + 0.8 KV	lossless
FP16	16	1384.7 GB 1383.9 + 0.8 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 ~207 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 A.X K2 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)

AIME97.1

GPQA Diamond85.7

LiveCodeBench84.0

AA Long Context70.0

SciCode41.0

Terminal-Bench 2.139.0

HLE29.6

Terminal-Bench 40.0

Run this model

▸Easiest way to get started·Beginners

DOCS ↗

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

$ollama run other:692b-q4_K_M

Tag may need adjustment — check ollama.com/library/other 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 A.X K2

Build Hardware for A.X K2

QUANT	BPW	VRAM	QUALITY
IQ2_XXS	2.38	206.3 GB	65%
IQ2_M	2.93	253.8 GB	75%
Q2_K	3.16	273.7 GB	78%
IQ3_XXS	3.25	281.5 GB	82%
IQ3_XS	3.5	303.1 GB	84%
Q3_K_S	3.64	315.2 GB	85%
IQ3_M	3.76	325.6 GB	86%
Q3_K_M	4	346.3 GB	88%
Q3_K_L	4.3	372.3 GB	90%
IQ4_XS	4.46	386.1 GB	92%
Q4_K_S	4.67	404.3 GB	93%
Q4_K_M	4.89	423.3 GB	94%
Q5_K_S	5.57	482.1 GB	96%
Q5_K_M	5.7	493.3 GB	96%
Q6_K	6.56	567.7 GB	97%
Q8_0	8.5	735.4 GB	100%
FP16	16	1383.9 GB	100%