LG K-EXAONE 2.0 Card - ROK-Fortress: leaderboard
ROK-Fortress adversarial prompts in English and Korean with US or Korean context, judged by a Kimi K2 Thinking, o4-mini and GPT-5.2 ensemble; score is 100 minus the mean TRS risk score, so higher is safer. LG AI Research K-EXAONE 2.0 model card main table (July 2026), which its technical report (arXiv 2608.04505, Table 6) gives as reasoning mode for all five models.
Metric: Safety Score (100 - mean TRS). Source: huggingface.co. Saturation forecast: Around April 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 397B A17B (Reasoning) | 86.1 | #154 (Qwen 3.5 397B A17B) |
| 2 | GLM-5.1 (Thinking) | 73.2 | #109 (GLM-5.1) |
| 3 | K-EXAONE (Reasoning) | 60.9 | #387 (K-EXAONE) |
| 4 | DeepSeek V4 Pro (Max) | 47.6 | #112 (DeepSeek V4 Pro) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=lg-k-exaone-2-0-card-rok-fortress · How It Works · Data refreshed daily, snapshot 2026-11-05.