A.X-K2: benchmark results
SK Telecom's open 688B-total/33B-active Korean-English agentic MoE, with unified thinking and non-thinking modes and a 256K context. Provider: SK Telecom. Released 2026-07-29. Access: Open.
Aggregate Rating 1659 ± 20, rank #293 of 1656 rated models, from 177 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SKT A.X K2 Card - CLIcK | 91.6 | Accuracy (%) | 100 |
| SKT A.X K2 Card - KMMLU-Pro | 80.5 | Accuracy (%) | 100 |
| AA-Omniscience Hallucination Rate | 32.99 | Hallucination Rate (%) | 90.2 |
| Nejumi 4 - jaster (2-shot) - JSICK | 84 | Exact match (%) | 89.5 |
| Nejumi 4 - jaster (2-shot) - JCoLA (in-domain) | 81 | Exact match (%) | 88.8 |
| SKT A.X K2 Card - KMO 2026 (1st Round) | 92.5 | Accuracy (%) | 85.7 |
| SKT A.X K2 Card - KoBALT | 73 | Accuracy (%) | 85.7 |
| Nejumi 4 - jaster (2-shot) - JaNLI | 100 | Exact match (%) | 85 |
| AA GPQA Diamond | 85.66 | Accuracy (%) | 79.4 |
| Nejumi 4 - MT-Bench (Japanese) - Writing | 97.5 | Judge rating (1-10, x10) | 78.3 |
| AA-Omniscience Index - Law | -4 | Omniscience Index | 77.5 |
| AA CritPt | 8.86 | Accuracy (%) | 77.4 |
Interactive version: theaggregate.ai/model?slug=a-x-k2 · How It Works · Data refreshed daily, snapshot 2026-10-20.