GPT-5.6-Cyber

Why teams choose it

Tradeoffs to know

• Not a general consumer model—expect vetting, hardware security keys, monitoring, and approved-use restrictions.
• Still dual-use capable; misuse risk remains high without organizational controls and human authorization.
• No public list price in the announcement—access is program-gated rather than self-serve token pricing.

When not to use this

• OpenAI does not position Cyber as Critical-capability; do not treat vendor framing as a compliance sign-off for your environment.
• Not ideal for simple tasks where cheaper models in the same lineup are good enough.
• Avoid for regulated or high-stakes outputs without evaluations that mimic your tooling, data, and review process.

Technical specs

Inputs

text

Outputs

text

Capabilities

authorized vulnerability research, exploit validation support, red-team assistance, reduced dual-use refusals for approved defenders, security testing support, defensive cyber operations assistance

License

Proprietary

Model string

gpt-5-6-cyber

Benchmarks

{
  "notes": "OpenAI reports GPT-5.6-Cyber reaches High (not Critical) cyber capability under its Preparedness Framework. Announcement materials cite roughly 95% completion on requested advanced cybersecurity tasks in internal testing versus low single-digit rates for standard GPT-5.6 Sol safeguard settings—treat as vendor-reported, not an independent GenAIWiki benchmark.",
  "access": "Daybreak Red only (vetted)",
  "source": "https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/",
  "daybreak_role": "Daybreak Red purpose-trained cyber model",
  "vendor_reported": true,
  "preparedness_cyber": "High (below Critical)"
}