OpenAI released GPT-6 Cyber on November 5, adding a specialized model for investigating software vulnerabilities, reproducing them in authorized test environments and checking whether fixes work. The focus is practical: moving from a suspected bug to a verified repair. The release is listed in the official release notes.
Key details
- Large context: 1.05 million tokens, with standard short-context pricing of US$20 per million input tokens and US$100 per million output tokens. See the model specifications.
- Restricted availability: Daybreak Red access plus separate model approval are required. API deployment also needs project access and IP allowlisting. See the access requirements.
- Higher capability classification: OpenAI rates GPT-6 Cyber as “Critical” for cybersecurity, versus “High” for GPT-5.6 Cyber, under its Preparedness Framework. This is OpenAI’s assessment, not an independent performance ranking. See the official cybersecurity documentation.
Why it matters
For teams building apps, websites and developer tools, the opportunity is more thorough security investigation before a release, not simply another code-writing assistant.
For SunMarc App Labs, the useful evaluation would be whether it improves vulnerability reproduction and patch verification enough to justify its access requirements and cost. The release alone does not establish that benefit.
For another specialized model with restricted access, see our coverage of GPT-6 Rosalind for life sciences research.