whatmodel.sh / A.X K1 (519B-A33B)

A.X K1 (519B-A33B)

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

A.X K1 (519B-A33B) is an open-weights language model from SK Telecom, released December 2025. Its published size is 518.98B parameters (33B active per token). Licence: apache-2.0.

Catalogue identity notes

skt/A.X-K1 @bb56f50 (created 2025-12-29, licence apache-2.0). SK Telecom A.X K1 (Dec 2025), 519B MoE with 33B active; predecessor of A.X K2. Lab rows from the A.X K2 technical report tables.

Model information sources

Release date ↗ · Weights licence ↗

HF safetensors metadata 518,983,059,456 parameters; 33B active per the maker card.

62.1±1.3#177 of 719 LLMsfrom 7 benchmark results across 1 source

$ how it compares

The LLMs ranked either side of A.X K1 (519B-A33B), out of 828 with a score, then more from SK Telecom.

#modelscorewhereRAMreleased
206Devstral 2 (2512) Mistral62±2local~75GB2025-12
207Gemini 2.5 Flash Preview (05-20) Google62±1comparison only2025-05
208A.X K1 (519B-A33B) SK Telecom62±1local~1091GB2025-12
209gpt-oss-120b OpenAI62±1api $0.26/M +local~67GB2025-08
210GPT-5 nano OpenAI62±1api $0.14/M2025-08
more from SK Telecom
101A.X K2 (688B-A33B) SK Telecom69±1local~1454GB2026-07

see all LLMs · find the ones that fit your machine

$ what it's good at

coding▁▁▁▃▆██▇▆▄▃▁ #151 of 458 * · 15% of its score · 2 tests

Beats Gemini 2.5 Pro, Gemini 3.1 Flash-Lite, Qwen3-Coder-Next (80B-A3B) here, though they rank higher overall. Trails Ministral 3 8B Reasoning 2512, Phi-4-reasoning 14B, DeepSeek V3 (Dec '24) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LiveCodeBench v6 (Feb–May 2025) lab74.9% ✎▄··█····▄▄·█ beats 1/6medium · avg 45%▇▇▇▇ 9.0%at its level
SciCode (level unspecified) lab24.1% ✎▄·····▄█▄▄·▄ beats 0/6hard · avg 16%▇▇ 6.1%at its level
math▁▂▃▅▆▅▄█▄▁▁▁ #122 of 398 * · 11% of its score · 1 test

Beats Claude Sonnet 4.6, MiniMax-M3, MiMo-V2.5 (310B-A15B) here, though they rank higher overall. Trails Olmo 3.1 32B Think, Nemotron-Cascade-14B-Thinking, Gemini 2.5 Flash here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
AIME 2026 lab91.3% ✎▄·▄·▄▄··█··▄ beats 2/6medium · avg 56%▇▇▇▇ 11%sharp · noisy
other tests4 not in a task · 74% of its score
testresultvs the others in its rundifficultycountswhy
GPQA Diamond lab76.4% ✎▄·····▄█▄▄·▄ beats 0/6hard · avg 45%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 38%sharp · at its level
Global-MMLU-Lite lab77.8% ✎▃······▅··█▃ beats 0/6medium · avg 60%▇▇▇▇▇▇▇ 18%at its level
IFBench (aggregation unspecified) lab63.2% ✎▃······█▃▃·▃ beats 0/6medium · avg 44%▇▇▇▇ 8.9%at its level
Humanity's Last Exam lab8.3% ✎▃········█·▂ beats 0/6very hard · avg 5.4%▇▇▇ 8.2%typical

$ where to run it

Other ways to run it

  1. Run it as published · needs ~1038 GB, NVIDIA only · from huggingface.co/skt
    vllm serve skt/A.X-K1 --tensor-parallel-size 8

Weights and documentation: huggingface.co/skt ↗

$ how the score was worked out · A.X K2 Technical Report (SK T… 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

sourceresultsshare of the scorelicence
A.X K2 Technical Report (SK Telecom), arXiv 2608.30181v1 · lab-reported7▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 99%CC BY 4.0
prior · pull toward the middle▇ 1%
weighted average= 62.1

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

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