OpenAI reports that GPT-6 Astra significantly reduces factual errors and user-reported hallucinations compared to GPT-5.6 Sol, while also achieving a 'Critical' cybersecurity capability threshold for heightened safety in high-risk scenarios.
Open the full topic