MiniMax M3.1 Flash
A multimodal model with downloadable weights for coding, visual understanding and long-context agent workflows.
GPT-6 Cyber
openai.com
OpenAI's restricted cybersecurity model helps approved researchers discover vulnerabilities, validate exploits and develop patches within authorized scope.
Visit websitePricing
Paid
Available on
API, Codex
Category
Llm Models
What is GPT-6 Cyber?
GPT-6 Cyber is OpenAI's specialist cybersecurity model for finding and validating exploitable vulnerabilities within a researcher's authorized scope. Its work includes vulnerability investigation, exploit development, authorized security testing and candidate patches. It serves approved security researchers and teams conducting advanced security work.
Initial access is limited to selected approved Daybreak Red customers through the API and Codex. API integrations use the model identifier gpt-6-cyber. Access requires approval and project setup.
Key features
Vulnerability discovery and validation
Investigate software weaknesses and assess exploitability within an authorized engagement.
Patch development
Develop candidate fixes alongside vulnerability investigation and security testing.
Large context window
Supports a 1,050,000-token context window and up to 128,000 output tokens.
Text and image inputs
Accepts text and images, with text output. Audio and video modalities are unsupported.
Structured integrations
Supports streaming, function calling and structured outputs for application workflows.
Responses API tools
Supports hosted shell, apply patch, code interpreter, file search, web search, computer use and MCP through the Responses API.
Who is it for?
Application security teams investigating findings
Examine a suspected vulnerability and develop a candidate patch within an approved assessment.
Researchers validating exploitability
Evaluate whether a weakness can be reproduced in an authorized test environment.
Developers integrating security workflows
Build an internal application that returns structured investigation results for review.
How to use GPT-6 Cyber
- Confirm Daybreak Red access and separate GPT-6 Cyber approval.
- Enable access for the API project; use a non-default project for project-level controls.
- Configure an IP allowlist for the application's outgoing public IP addresses.
- Use the approved project's key and select gpt-6-cyber with the daybreak_red access program in Responses API requests.
- If the project uses Zero Data Retention, configure Private Safety Processing.
GPT-6 Cyber pricing
Paid
API usage is metered, with separate input, cached-input, cache-write and output rates. Consult the official API pricing page for current rates. Model approval is required independently of billing.
Check current pricing ↗What to know before you use it
- Daybreak Blue approval does not grant specialist-model access. GPT-6 Cyber requires Daybreak Red access and separate model approval.
- Fine-tuning is unsupported.
- Automated API safeguards can temporarily limit access during review, including occasional flags on legitimate defensive research. Codex uses different safeguards.
- Tool support does not automatically configure an execution environment or grant access to external systems.
Frequently asked questions
Can any API customer use GPT-6 Cyber?
No. Organizations need Daybreak Red access and specific approval for GPT-6 Cyber.
Is it available in Codex?
Yes, for selected approved Daybreak Red customers. Initial availability also includes API access.
Does Daybreak approval include Zero Data Retention?
No. Retention controls require separate confirmation. Projects using Zero Data Retention with GPT-6 Cyber must configure Private Safety Processing.