squ11z1/Mythoseek
Mythoseek is a 9 billion parameter language model developed by Merlin Research, fine-tuned from DeepSeek V4 Pro-Qwen3.5. It specializes in cybersecurity tasks such as vulnerability research, penetration testing, OSINT, and CWE-pattern reasoning. This model was distilled from enterprise pentest reports and frontier model traces, offering closed-source cyber AI capabilities to the open community. It is particularly strong in identifying real-world vulnerabilities and assisting with security report generation.
Loading preview...
Mythoseek: A Cybersecurity-Specialized LLM
Mythoseek is a 9 billion parameter language model developed by Merlin Research as part of their KAON quantum-classical research program. It is fine-tuned from DeepSeek V4 Pro-Qwen3.5 and specifically designed for cybersecurity applications, including vulnerability research, penetration testing, OSINT, and CWE-pattern reasoning. The model was trained using a multi-stage pipeline involving SFT distillation from frontier models, GRPO/RL with verifiable rewards on cyber tasks, tool-use SFT, and CWE-pattern structured reasoning.
Key Capabilities & Performance
- Cybersecurity Specialization: Excels in tasks related to vulnerability analysis, penetration testing assistance (OSINT, recon, XSS, SQLi), and CWE classification.
- Quantum-Classical Integration: Its training incorporated OTOC scrambling measurements from real IBM Quantum hardware to calibrate Adaptive Entropy Regularization (AER) coefficients during GRPO training.
- Benchmark Performance: Achieved 62% pass@4 on Level 0 (Full scaffolding) of the CyberGym benchmark, which evaluates against 1,507 real-world vulnerabilities. This positions it as a strong open-weight alternative to closed models like Claude Mythos Preview.
Intended Use Cases
- Vulnerability research and CVE analysis
- Penetration testing assistance
- CWE classification and pattern recognition
- Security report generation
- Red team reasoning support