OpenAI’s developer changelog lists GPT-6-Cyber on November 5, 2026. It is a restricted model for security work, not a general release.
What to know
- What OpenAI says it does: investigate vulnerabilities, reproduce them in an authorised test environment and verify fixes. Model ID:
gpt-6-cyber. - Access: OpenAI’s model page says an organisation needs Daybreak Red access and approval to use it. Its Daybreak guide adds that GPT-6-Cyber needs Daybreak Red plus additional model approval, confirmed through your account team. Selecting a program in a request does not grant access.
- Setup requirements in the guide: SSO and MFA for the organisation, project access to the model, an IP allowlist, and a stable pseudonymous
safety_identifierper user. - Specs on the model page: text and image input, text output, a 1,050,000-token context window and 128,000-token maximum output. Only the Chat Completions and Responses endpoints are supported.
- Price per million tokens, per OpenAI: $20 input, $2 cached input and $100 output.
Why it matters
Gated models are becoming a pattern: OpenAI also limits GPT-6 Rosalind to approved organisations. Most readers cannot use GPT-6-Cyber, so treat it as a signal about where access-controlled AI is heading. For models you can use today, see the ChatGPT page and the model tracker. We did not read OpenAI’s announcement page (blocked for us), so we report only what the API docs state.
Source: OpenAI API docs. We summarise in our own words; check the source for the full details.