GPT-6-Cyber model (API docs)

Summary

OpenAI released GPT-6-Cyber on 2026-11-05, a cybersecurity model for investigating vulnerabilities, reproducing them in an authorized test environment, and verifying fixes. It is limited to organizations approved for OpenAI's Daybreak Red programme. The announcement post ("GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents") returned 403 to automated fetches, so this note uses the API model page and changelog.

Key points

• Model ID: gpt-6-cyber; Chat Completions and Responses APIs only.
• Access: requires Daybreak Red access and organization approval; set up the organization and project with the Daybreak setup guide before the first request.
• Limits: 1,050,000-token context (922,000 max input), 128,000 max output; text and image in, text out.
• Price (2026-11-05, per 1M tokens): $20 input, $2 cached input, $25 (30-minute) or $50 (12-hour) cache writes, $100 output.
• Tools: function calling, web search, file search, tool search, code interpreter, hosted shell, apply patch, skills, computer use, MCP, structured outputs, streaming.
• Earlier cyber models this year (GPT-5.4, 5.5, and 5.6 Cyber) come from secondary reports only and were not checked.
• Benchmarks and safeguards are described in the announcement post, which could not be read.

Why it matters

OpenAI's security-focused model is available only after vetting, not to every API customer, so most teams cannot use it directly.