ACI_Cyber_Base_GPT_OSS_20B is an LLM derivied from openai/gpt-oss:20b, continued pretrained with security related corpus including computer security QA, CVE descriptions, MITRE ATTCK framework descriptions, Incident Response playbooks, and penetration testing knowledge QA.

Multi-judge experiment demonstrates that this model consistently outperforms openai/gpt-oss:20b in SOC investigation planning:

Detailed experiment is documented here

This gpt_oss model was trained 2x faster with Unsloth and Huggingface's TRL library.