LLM
A.X K2
SK Telecom 688B foundation model trained from scratch, released on Hugging Face, with a focus on Korean knowledge and long-context reasoning.
Levelmodel
ClassLLM
Parameters688B
Context262,144
Released29 Jul 2026
Opennessopen weights permissive
Runs onnot in the notes
Statusactive
Benchmark Results
vendor-reported
AA-LCR v1.1 v1.1long context
Thinking mode; AA-LCRvendor-reported
66
050100
Score in %. Higher is better.
A.X K2 model card
BrowseCompagents and tool use
Thinking mode; At most 10 searchesvendor-reported
9.3
050100
Score in %. Higher is better.
A.X K2 model card
GPQA Diamondknowledge and reasoning
Thinking modevendor-reported
85.6
050100
Score in %. Higher is better.
A.X K2 model card
Humanity's Last Exam (HLE)knowledge and reasoning
Thinking modevendor-reported
27.8
050100
Score in %. Higher is better.
A.X K2 model card
IFBenchknowledge and reasoning
Thinking modevendor-reported
75.9
050100
Score in %. Higher is better.
A.X K2 model card
KMMLU-Proknowledge and reasoning
Thinking modevendor-reported
80.5
050100
Score in %. Higher is better.
A.X K2 model card
LiveCodeBench v6coding
Thinking mode; Feb to May windowvendor-reported
84
050100
Score in %. Higher is better.
A.X K2 model card
RULERlong context
Thinking mode; Overall averagevendor-reported
94.6
050100
Score in %. Higher is better.
A.X K2 model card
SciCodeknowledge and reasoning
Thinking modevendor-reported
41
050100
Score in %. Higher is better.
A.X K2 model card
tau2-benchagents and tool use
Thinking mode; Telecomvendor-reported
98
050100
Score in %. Higher is better.
A.X K2 model card
Terminal-Bench 2.1agents and tool use
Thinking modevendor-reported
36
050100
Score in %. Higher is better.
A.X K2 model card
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).
R&D Topics
Identity
- Developer
- SK Telecom
- Family
- A.X
- Released
- 29 Jul 2026 sourcesannouncement page
- Weights published
- –
- Record checked
- 11 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.
Architecture and Size
- Architecture
- MoE Decoder-only MoE, 61 layers, 256 routed experts plus 1 shared (8 active), Multi-head Latent Attention with Sparse Gated Attention
- Total parameters
- 688B sources
- Active parameters
- 33B
- Checkpoint parameters
- –
- Context length
- 262,144 tokens
- Input
- text
- Output
- text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).
Openness and Access
- Openness
- open weights permissive
- Licence
- Apache License 2.0 sources
- Access routes
- open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.
Deployment
- Runs on
- not in the notes
- Footprint
- not in the notes
- Runtimes
- not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.
Indonesia
- Indonesian support
- not stated
- Regional languages
- not in the notes
- Jakarta hosting
- unknown
- Languages
- English, Korean, Chinese, Japanese, Spanish
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.
Lineage
The notes record no lineage for this model.
Industry Use Cases
No industry use case links this model.
Links
DeveloperSK Telecombig tech, South Korea
SK Telecombig tech, South KoreaModelA.X K2SK Telecom, LLM
A.X K2SK Telecom, LLMR&D TopicsFrontier Pretraining ScalingRegional or National Foundation Model Training (Continued Pretraining at 7B to 70B)
Use CasesIndustrial Copilots and Maintenance AssistantsIndustrialIndonesian and Regional-Language LLMsPublic SectorClinical LLM (Frontier Proprietary)HealthcarePharma Documents, Regulatory and Life-Science R&DHealthcare+6 moreUse cases of models in the same R&D topic; none is recorded for this model itself.
Who Runs It
PT Pertamina Geothermal EnergyIn production, 14 Jul 2026
PT PegadaianIn production, 5 Jun 2026
Bank Central Asia (BCA)In production, 16 Mar 2026Bank Central Asia (BCA)In production, 16 Mar 2026+51 moreDeployments of these use cases; none names this model.
PT Pertamina Geothermal EnergyIn production, 14 Jul 2026PT PegadaianIn production, 5 Jun 2026Bank Central Asia (BCA)In production, 16 Mar 2026Bank Central Asia (BCA)In production, 16 Mar 2026+51 moreDeployments of these use cases; none names this model.Models 1
Organisations 1
R&D topics 2
Frontier Pretraining ScalingTraining a new base model at or near the top-5 compute of its release date (Epoch's front…produced byRegional or National Foundation Model Training (Continued Pretraining at 7B to 70B)Pretraining or continued pretraining a mid-size model to serve a national or regional lan…produced by
Sources
- news.sktelecom.com/en/3204 https://news.sktelecom.com/en/3204
- huggingface.co/skt/A.X-K2 https://huggingface.co/skt/A.X-K2
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "A.X K2", https://aurion.semai.id/models/sk-telecom-a-x-k2/, records of October 2026.