OpenAegis

OpenAegis is a cybersecurity-specialized large language model introduced in the paper "CyberFactory: Scaling Cyber Security Capabilities with Instances from the Wild" (arXiv:2608.23181). It is produced via full-parameter supervised fine-tuning (SFT, three epochs) of the Qwen 3.5-397B-A17B base model (a mixture-of-experts model with 397B total and 17B active parameters).

The model is trained on trajectories generated by CyberFactory, an open-source framework that connects data construction, trajectory synthesis, and model training across three tasks: proof-of-concept (PoC) generation, vulnerability patching, and cybersecurity question answering (CyberQA). On the CyberGym benchmark, OpenAegis reaches 58.1% Pass@1 under a one-hour per-task budget, improving over its Qwen 3.5 base by 28.5 points.