NEOS is a 27B model fine-tuned on real security data. It reasons through attack surfaces and executes exploits autonomously.
Capabilities
NEOS doesn't suggest commands. It executes them, reads the output, adapts, and retries until the objective is reached.
Performance
500 cybersecurity MCQ. NEOS v15: 89% — surpassing GPT-4o at 40% fewer parameters.
2,500 threat intelligence questions. 73% — a +24% jump from v14 after ATT&CK fine-tuning.
8 real-world tasks with CVE lookup, MITRE ATT&CK, and web search. 9.8/10 average.
Pricing
Every team is reviewed and approved manually. No self-service signups — we onboard intentionally.
Individual researchers and small red teams.
Pentest firms running multiple engagements.
MSSPs and orgs needing on-premise deployment.
Contact
NEOS is in closed early access. We're onboarding red teams and security researchers before public launch.
Penetration testing firms, enterprise red teams, MSSPs, and serious security researchers only.
We'll review your request within 24 hours.