OpenAI Introduces GPT-6 Cyber Model

OpenAI introduced GPT-6 Cyber, a cybersecurity model trained to find and validate exploitable vulnerabilities while remaining within research boundaries. OpenAI states the model is classified as Critical under its Preparedness Framework, an increase from the High classification assigned to GPT-5.6 Cyber. The model is integrated into the Daybreak program, where Daybreak Red enables authorized security testing and exploit development.

On benchmark evaluations, GPT-6 Cyber achieves a 48.5% score on ExploitGym, up from 32.1% for GPT-5.6 Cyber. On an offensive web penetration testing evaluation, GPT-6 Cyber attained 40.3% recall compared with 32.0% for GPT-6 Astra. When tested on RemoteExploitBench, GPT-6 Cyber completed 95% of tasks, while GPT-6 Astra running with Daybreak Blue refused 100% of tasks. Current model pricing is provided in the OpenAI API documentation.

According to OpenAI, alignment evaluations show improvements over prior versions. On the ExploitGym Honeypot benchmark, GPT-6 Cyber stayed within bounds 99.93% of the time without deployment safety prompts and 100% of the time with prompts, compared to 94.49% for GPT-5.6 Cyber.

In real-world testing, researchers used the model to identify 81 novel high-severity vulnerabilities in Firefox and developed 948 privilege escalation exploits in Linux out of 1,724 reproduced vulnerabilities. This compares to 166 exploited by GPT-5.6 Sol using the same research harness. Security teams at Cisco, Calif, CodeRabbit, Cognition, Palo Alto Networks, SentinelOne, and Perplexity are using GPT-6 Cyber for security research.