Imperum CybersecurityLLM v1.0 brings specialised AI to security teams, with free downloads and commercial use under Apache 2.0.
Amsterdam, August 24, 2026. Imperum has released Imperum CybersecurityLLM v1.0, a language model built for cybersecurity work and designed to run inside an organisation’s own environment. Developed in partnership with Alican Kiraz, the model is available to the security community on Hugging Face. The release gives security teams a way to use specialised AI while keeping their prompts, investigation context and security data on infrastructure they control.
Built for the work security teams do
Fine-tuned on curated cybersecurity instruction data, Imperum CybersecurityLLM covers SOC and SIEM operations, detection engineering, digital forensics and incident response, malware analysis, threat intelligence, cloud security, and OT and ICS environments. It can help analysts investigate alerts, draft detection logic, explain suspicious activity and turn technical findings into clearer assessments. Outputs are intended to support analyst review and decision-making.
Local deployment, practical access
The model supports fully local and air-gapped deployment, with no external per-token API charges for local inference. It runs with llama.cpp, Ollama and LM Studio, and can connect to existing workflows through an OpenAI-compatible API. With approximately 35 billion total parameters and around 3 billion active per token, the model achieved approximately 50 tokens per second for a single stream during testing on one NVIDIA DGX Spark. Performance varies with hardware and configuration.
Available to download
Imperum CybersecurityLLM v1.0 is available in 4-bit and 8-bit GGUF formats under Apache 2.0, including permission for commercial use. The Hugging Face model card provides setup instructions, technical details and guidance on known limitations. We invite security practitioners to try the model, share feedback and tell us where it needs to improve.