G
GPT-6 Cyber
OpenAI · models
OpenAI's cybersecurity-specialised model, built to find and validate exploitable software vulnerabilities. It is the fourth cybersecurity-focused system OpenAI has shipped in 2026.
Specifications
| Developer | OpenAI |
|---|---|
| Type | Cybersecurity LLM |
| Speciality | Vulnerability discovery and exploitation |
| Context | Not disclosed |
| Modality | Text |
| Access | API, ChatGPT (cybersecurity programs) |
Pricing
Available through OpenAI's cybersecurity programs; pricing not publicly listed.
Strengths
- 48.5% on ExploitGym, up from 32.1% for GPT-5.6 Cyber
- Designed to stay within researcher-set boundaries
- Backed by OpenAI's safety and red-teaming infrastructure
Weaknesses
- Restricted availability
- Specialised — not a general-purpose model
- High compute requirements for exploitation tasks