Independent evaluations will determine whether SK Telecom’s 688B-total, 33B-active A.X-K2 release delivers competitive multilingual capability at practical inference costs for an open-weight model.
state: expiredheat: lowuncertainty: highnovelscott: nonea-x-k2 south-korea-ai open-models local-inferenceSK TelecomKRAFTON
What is this?
The case describes A.X-K2 as an open-weight SK Telecom model with 688B total parameters and 33B active parameters, positioned for multilingual capability and economical inference. Supplied snippets establish that SK Telecom previously open-sourced A.X models, has emphasized Korean-language efficiency and production inference costs, and planned A.X-K2 as a five-language follow-up to A.X-K1. However, the snippets do not independently verify the reported K2 release, its parameter counts, KRAFTON’s role, benchmark performance, licensing terms, or practical serving costs, so those claims remain pending independent evaluation.
Why it matters to Scott
No intersection found in Scott’s wikis, and no radar pages indicate this release or its actors are already tracked. The claimed capabilities and inference economics also remain unverified, so the supplied material does not yet establish a consequential connection.
queries asked of Scott's wikis
- open-weight model sovereignty and national AI strategy
- Mixture-of-Experts active parameters versus inference economics
- multilingual model evaluation beyond vendor benchmarks
- self-hosted frontier models and practical serving costs
- local inference hardware requirements and deployment tradeoffs
- open-weight releases versus reproducible independent evaluations
Measured heat
no measured readings yet — the hourly heat pass fills this in
How the heat travelled
no chain yet — the hourly chain pass fills this in
Evidence (2) — ⭐ canonical anchor
Interpretation history
2026-07-29T08:29:52Z
No independent evaluation, deployment, or inference-cost evidence has emerged since creation. The case has been reobserved eight times with no new signal — the release is real but its practical significance remains entirely unvalidated, and the window for meaningful corroboration has passed without materializing.
2026-07-29T07:28:27Z
The evidence remains limited to the first-party release artifact and discussion that does not test multilingual quality, deployability, or inference economics. Repeated reobservation is adding no meaning; pause until independent benchmarks or implementation reports emerge.
2026-07-29T06:26:55Z
No independent evaluation, implementation, or serving-cost evidence has emerged; the update is another reobservation of the release artifact and speculative discussion. Repeated checks are no longer adding meaning, so the case remains a cold seed pending substantive third-party testing.
2026-07-29T05:24:01Z
The newly attached material adds no independent benchmark, deployment, or serving-cost evidence; it remains repetition of the release artifact and speculative discussion. The model’s practical competitiveness is still entirely unvalidated despite broader interest in open and local models.
2026-07-29T04:24:41Z
The attached material still only confirms the first-party release and architecture claims; no independent benchmark, deployment, or inference-cost evidence changes the hypothesis. Attention remains modest and repetitive rather than corroborating.
2026-07-29T03:22:02Z
The first-party release artifact confirms the model’s existence and stated architecture, but no independent evaluation or implementation evidence has arrived to validate capability or inference economics. Discussion and engagement are static, so the case remains an uncorroborated release awaiting substantive testing.
2026-07-29T02:24:15Z
grounded: novel/none — No intersection found in Scott’s wikis, and no radar pages indicate this release or its actors are already tracked. The claimed capabilities and inference econo
2026-07-29T02:23:32Z
origin walked (codex/luna, conf 0.95): anchor reddit.post.1v9hpac -> echo.github.99e5f27236 by SK Telecom (SKT-AI; commit by Kim, Jin)
2026-07-29T02:21:40Z
case created — A major open-weight MoE release associated with South Korea’s sovereign-AI program merits focused performance and deployability validation.
Decision trace
- 07-29 18:29expireNo independent evaluation, deployment, or inference-cost evidence has emerged since creation. The case has been reobserved eight times with no new signal — the release is real but its practical signif
- 07-29 18:20mark_dirtyengagement_update
- 07-29 17:28repriceThe evidence remains limited to the first-party release artifact and discussion that does not test multilingual quality, deployability, or inference economics. Repeated reobservation is adding no mean
- 07-29 17:20mark_dirtyengagement_update
- 07-29 16:26repriceNo independent evaluation, implementation, or serving-cost evidence has emerged; the update is another reobservation of the release artifact and speculative discussion. Repeated checks are no longer a
- 07-29 16:20mark_dirtyengagement_update
- 07-29 15:24repriceThe newly attached material adds no independent benchmark, deployment, or serving-cost evidence; it remains repetition of the release artifact and speculative discussion. The model’s practical competi
- 07-29 15:20mark_dirtyengagement_update
- 07-29 14:24repriceThe attached material still only confirms the first-party release and architecture claims; no independent benchmark, deployment, or inference-cost evidence changes the hypothesis. Attention remains mo
- 07-29 14:20mark_dirtyengagement_update
- 07-29 13:22repriceThe first-party release artifact confirms the model’s existence and stated architecture, but no independent evaluation or implementation evidence has arrived to validate capability or inference econom
- 07-29 13:20mark_dirtyengagement_update
- 07-29 12:24groundNo intersection found in Scott’s wikis, and no radar pages indicate this release or its actors are already tracked. The claimed capabilities and inference economics also remain unverified, so the supp
- 07-29 12:23promote_anchororigin walk conf 0.95
- 07-29 12:21createA major open-weight MoE release associated with South Korea’s sovereign-AI program merits focused performance and deployability validation.