GPT Cyber 6

Release Summary

OpenAI cybersecurity model for approved Daybreak Red customers, available first in the API and Codex. API ID gpt-6-cyber. Short-context rates are $20/M input, $2/M cached input, $25/M cache writes for 30 minutes ($50/M for 12 hours), and $100/M output. Long-context rates are not published, and Batch is not supported. Context is 1.05M tokens, with 922k max input and 128k max output. Text and image in, text out. OpenAI reports 48.5% on ExploitGym, compared with 32.1% for GPT-5.6 Cyber, and 40.3% recall on its web-vulnerability evaluation, compared with 32.0% for GPT Astra 6. On ExploitGym Honeypot, in-scope behavior is 99.93% without the deployment prompt and 100% with it, compared with 94.49% for GPT-5.6 Cyber. OpenAI classifies the model as Critical for cybersecurity under its Preparedness Framework. Access widens later to all Daybreak Red customers.

Timeline