What changed
OpenAI's November 5 developer changelog lists GPT-6-Cyber, a model for investigating security flaws and checking fixes in authorized test environments. Developers request it through the API, the interface software uses to send requests. The model ID is gpt-6-cyber. [1] [2]
OpenAI's announcement says the first group can use the model through the API and Codex, its coding assistant. It says access will expand to more Daybreak Red customers. [4]
Check the project before you plan a test
OpenAI's setup guide says the organization needs Daybreak Red and approval for this specific model. An organization owner enables Daybreak Red for the project that sends requests. The guide also requires an IP allowlist, a list of public network addresses from which your app can send requests. [3]
OpenAI's standard short-context rates are $20 per million input tokens and $100 per million output tokens. Tokens are the small chunks of text used to measure how much you send and receive. Check the full pricing details for your workload before estimating a run. [2]
Our take
Give the model one narrow question, such as what a test should check after a fix. Use a test copy and sample data. Write down the systems and accounts in scope before anyone runs a check, and ask a colleague to review what the model suggested and what the test actually showed.
This briefing is based on the published model and setup documentation.