JuicesAI/CTFty-9b

CTFty-9b: A Specialized Security Agent Model

CTFty-9b is a 9 billion parameter language model developed by JuicesAI, specifically engineered as a security agent. It is a full fine-tune of the XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B base model, with its predecessor (v1) being trained from Qwen3.5-9B.

Key Capabilities

• CTF Challenge Solving: Trained on 42 verified agent runs from various CTF competitions (Kaspersky CTF, KubSTU, RCPISS/Avito, НТО), covering diverse categories like reverse engineering, pwn, forensics, web, and crypto.
• Pentesting Support: Incorporates knowledge from 535 pentest walkthroughs (Pentest-R1) to assist in penetration testing tasks.
• Cybersecurity Knowledge: Enhanced with 858 MITRE ATT&CK technique cards and 216 MBC malware behavior cards, providing a deep understanding of cybersecurity tactics and malware analysis.
• Specialized Fine-tuning: Underwent a full fine-tuning process over 2 epochs with a learning rate of 5e-6, optimized for its specific security domain.

Good For

• Automated CTF Participation: Ideal for developing agents that can autonomously tackle Capture The Flag challenges.
• Security Research and Analysis: Useful for researchers and analysts needing to understand and simulate various cybersecurity techniques.
• Pentesting Assistance: Can serve as a valuable tool for generating insights or automating aspects of penetration tests.

Usage

The model utilizes the same ChatML format as its base model, making it compatible with standard AutoModelForImageTextToText and AutoProcessor implementations.