XekRung: A Frontier LLM for Cybersecurity
A new large language model tailored for cybersecurity, named XekRung, has been unveiled by researchers. This model leverages a variety of data synthesis pipelines to facilitate the scalable creation of high-quality training datasets, establishing a robust base for cybersecurity expertise. Its capabilities are enhanced through a comprehensive training pipeline that includes continued pre-training (CPT), supervised fine-tuning (SFT), and reinforcement learning (RL). An evaluation system that assesses multiple dimensions aids in the ongoing refinement of both specialized and general skills. Tests indicate that XekRung outperforms other models of similar scale on cybersecurity benchmarks while also excelling in general assessments.
Key facts
- XekRung is a frontier large language model for cybersecurity.
- It uses diverse data synthesis pipelines for scalable training data construction.
- Training pipeline includes CPT, SFT, and RL.
- A multi-dimensional evaluation system is used for iterative improvement.
- State-of-the-art performance on cybersecurity benchmarks at same scale.
- Maintains strong performance on general benchmarks.
- arXiv:2605.00072v1
- Announce Type: cross
Entities
Institutions
- arXiv