OpenAI launches GPT-6 Cyber model with 48.5% score on ExploitGym

Cointime update: OpenAI released GPT-6 Cyber, a cybersecurity-focused model that scored 48.5% on the ExploitGym benchmark, ahead of GPT-6 Astra at 42.4% and GPT-5.6 Cyber at 32.1%, and helped uncover 81 previously unknown Firefox vulnerabilities. Out-of-bounds execution in testing fell to under 1% from about 5% in the prior generation, and the model adds abuse safeguards against high-risk attacks on critical infrastructure such as power plants and water facilities. GPT-6 Cyber is rated “critical” and is available to select Daybreak Red customers, with wider access in the coming weeks.